107.

A position detector 105, which may be a conventional encoder, may be operatively coupled to the conveyor belt 103 and the automation control system 108 to provide information corresponding to the movement (e.g., speed) of the conveyor belt 103. Thus, as will be further described herein, through the utilization of the controls to the conveyor belt drive motor 104 and/or the automation control system 108 (and alternatively including the position detector 105), as each of the material pieces 101 travelling on the conveyor belt 103 are identified, they can be tracked by location and time (relative to the various components of the system 100) so that the various components of the system 100 can be activated/deactivated as each material piece 101 passes within their vicinity. As a result, the automation control system 108 is able to track the location of each of the material pieces 101 while they travel along the conveyor belt 103.

In accordance with certain embodiments of the present disclosure, after the material pieces 101 are received by the conveyor belt 103, a tumbler and/or a vibrator may be utilized to separate the individual material pieces from a collection of material pieces, and then they may be positioned into one or more singulated (i.e., single file) streams. In accordance with alternative embodiments of the present disclosure, the material pieces may be positioned into one or more singulated (i.e., single file) streams, which may be performed by an active or passive singulator 106. An example of a passive singulator is further described in U.S. Pat. No. 10,207,296. As previously discussed, incorporation or use of a singulator is not required. Instead, the conveyor system (e.g., the conveyor belt 103) may simply convey a collection of material pieces, which have been deposited onto the conveyor belt 103 in a random manner.

Referring again to FIG. 1, certain embodiments of the present disclosure may utilize a vision, or optical recognition, system 110 and/or a distance measuring device 111 as a means to begin tracking each of the material pieces 101 as they travel on the conveyor belt 103. The vision system 110 may utilize one or more still or live action cameras 109 to note the position (i.e., location and timing) of each of the material pieces 101 on the moving conveyor belt 103. The vision system 110 may be further, or alternatively, configured to perform certain types of identification (e.g., classification) of all or a portion of the material pieces 101. For example, such a vision system 110 may be utilized to acquire information about each of the material pieces 101. For example, the vision system 110 may be configured (e.g., with a machine learning system) to collect any type of information that can be utilized within the system 100 to classify the material pieces 101 as a function of a set of one or more (user-defined) physical characteristics, including, but not limited to, color, hue, size, shape, texture, overall physical appearance, uniformity, composition, and/or manufacturing type of the material pieces 101. The vision system 110 captures images of each of the material pieces 101 (including one-dimensional, two-dimensional, three-dimensional, or holographic imaging), for example, by using an optical sensor as utilized in typical digital cameras and video equipment. Such images captured by the optical sensor are then stored in a memory device as image data. In accordance with embodiments of the present disclosure, such image data represents images captured within optical wavelengths of light (i.e., the wavelengths of light that are observable by the typical human eye). However, alternative embodiments of the present disclosure may utilize sensors that are able to capture an image of a material made up of wavelengths of light outside of the visual wavelengths of the typical human eye.

In accordance with certain embodiments of the present disclosure, one or more sensor systems 120 may be utilized solely or in combination with the vision system 110 to classify/identify material pieces 101. A sensor system 120 may be configured with any type of sensor technology, including sensors utilizing irradiated or reflected electromagnetic radiation (e.g., utilizing infrared (“IR”), Fourier Transform IR (“FTIR”), Forward-looking Infrared (“FLIR”), Very Near Infrared (“VNIR”), Near Infrared (“NIR”), Short Wavelength Infrared (“SWIR”), Long Wavelength Infrared (“LWIR”), Medium Wavelength Infrared (“MWIR”), X-Ray Transmission (“XRT”), Gamma Ray, Ultraviolet, X-Ray Fluorescence (“XRF”), Laser Induced Breakdown Spectroscopy (“LIBS”), Raman Spectroscopy, Anti-stokes Raman Spectroscopy, Gamma Spectroscopy, Hyperspectral Spectroscopy (e.g., any range beyond visible wavelengths), Acoustic Spectroscopy, NMR Spectroscopy, Microwave Spectroscopy, Terahertz Spectroscopy, including one-dimensional, two-dimensional, or three-dimensional imaging with any of the foregoing), or by any other type of sensor technology, including but not limited to, chemical or radioactive. Implementation of an XRF system (e.g., for use as a sensor system 120 herein) is further described in U.S. Pat. No. 10,207,296.

It should be noted that though FIG. 1 is illustrated with a combination of a vision system 110 and a sensor system 120, embodiments of the present disclosure may be implemented with any combination of sensor systems utilizing any of the sensor technologies disclosed herein, or any other sensor technologies currently available or developed in the future. Though FIG. 1 is illustrated as including a sensor system 120, implementation of such a sensor system is optional within certain embodiments of the present disclosure. Within certain embodiments of the present disclosure, a combination of both the vision system 110 and one or more sensor systems 120 may be used to classify the material pieces 101. Within certain embodiments of the present disclosure, any combination of one or more of the different sensor technologies disclosed herein may be used to classify the material pieces 101 without utilization of a vision system 110. Furthermore, embodiments of the present disclosure may include any combinations of one or more sensor systems and/or vision systems in which the outputs of such sensor and/or vision systems are utilized by a machine learning system (as further disclosed herein) in order to classify/identify materials from a heterogeneous mixture of materials, which can then be sorted from each other.

In accordance with alternative embodiments of the present disclosure, a vision system 110 and/or sensor system(s) may be configured to identify which of the material pieces 101 are not of the kind to be sorted by the system 100 (sometimes referred to as contaminants), and send a signal to reject such material pieces. In such a configuration, the identified material pieces 101 may be diverted/ejected utilizing one of the mechanisms as described hereinafter for physically moving sorted material pieces into individual bins.

Within certain embodiments of the present disclosure, the distance measuring device 111 and accompanying control system 112 may be utilized and configured to measure the sizes and/or shapes of each of the material pieces 101 as they pass within proximity of the distance measuring device 111, along with the position (i.e., location and timing) of each of the material pieces 101 on the moving conveyor belt 103. An exemplary operation of such a distance measuring device 111 and control system 112 is further described in U.S. Pat. No. 10,207,296. Alternatively, as previously disclosed, the vision system 110 may be utilized to track the position (i.e., location and timing) of each of the material pieces 101 on the moving conveyor belt 103.

Such a distance measuring device 111 may be implemented with a well-known visible light (e.g., laser light) system, which continuously measures a distance the light travels before being reflected back into a detector of the laser light system. As such, as each of the material pieces 101 passes within proximity of the device 111, it outputs a signal to the control system 112 indicating such distance measurements. Therefore, such a signal may substantially represent an intermittent series of pulses whereby the baseline of the signal is produced as a result of a measurement of the distance between the distance measuring device 111 and the conveyor belt 103 during those moments when a material piece 101 is not in the proximity of the device 111, while each pulse provides a measurement of the distance between the distance measuring device 111 and a material piece 101 passing by on the conveyor belt 103. Since the material pieces 101 may have irregular shapes, such a pulse signal may also occasionally have an irregular height. Nevertheless, each pulse signal generated by the distance measuring device 111 provides the height of portions of each of the material pieces 101 as they pass by on the conveyor belt 103. The length of each of such pulses also provides a measurement of a length of each of the material pieces 101 measured along a line substantially parallel to the direction of travel of the conveyor belt 103. It is this length measurement (and alternatively the height measurements) that may be utilized within certain embodiments of the present disclosure to determine when to activate and deactivate the acquisition of detected fluorescence (i.e., the XRF spectrum) of each of the material pieces 101 by a sensor system 120 implementing an XRF system so that the detected fluorescence is obtained substantially only from each of the material pieces and not from any background surfaces, such as a conveyor belt 103. This results in a more accurate detection and analysis of the fluorescence, and also saves time in the signal processing of the detected signals since only data associated with detected fluorescence from the material pieces is having to be processed.

Within certain embodiments of the present disclosure that implement sensor system(s) 120, the sensor system(s) 120 may be configured to assist the vision system 110 to identify the chemical composition, or relative chemical compositions, of each of the material pieces 101 as they pass within proximity of the sensor system(s) 120. The sensor system(s) 120 may include an energy emitting source 121, which may be powered by a power supply 122, for example, in order to stimulate a response from each of the material pieces 101.

Within certain embodiments of the present disclosure, as each material piece 101 passes within proximity to the emitting source 121, a sensor system 120 may emit an appropriate sensing signal towards the material piece 101. One or more detectors 124 may be positioned and configured to sense/detect one or more physical characteristics from the material piece 101 in a form appropriate for the type of utilized sensor technology. The one or more detectors 124 and the associated detector electronics 125 capture the received sensed characteristics to perform signal processing thereon and produce digitized information representing the sensed characteristics, which are then analyzed in accordance with certain embodiments of the present disclosure, and which may be used in order to classify (solely or in combination with the vision system 110) each of the material pieces 101. This classification, which may be performed within the computer system 107, may then be utilized by the automation control system 108 to activate one of the N (N≥1) sorting devices 126 . . . 129 for sorting (e.g., diverting/ejecting) the material pieces 101 into one or more N (N≥1) sorting bins 136 . . . 139 according to the determined classifications. Four sorting devices 126 . . . 129 and four sorting bins 136 . . . 139 associated with the sorting devices are illustrated in FIG. 1 as merely a non-limiting example.

The sorting devices may include any well-known mechanisms for redirecting selected material pieces 101 towards a desired location, including, but not limited to, diverting the material pieces 101 from the conveyor belt system into the plurality of sorting bins. For example, a sorting device may utilize air jets, with each of the air jets assigned to one or more of the classifications. When one of the air jets (e.g., 127) receives a signal from the automation control system 108, that air jet emits a stream of air that causes a material piece 101 to be diverted/ejected from the conveyor system 103 into a sorting bin (e.g., 137) corresponding to that air jet. High speed air valves from Mac Industries may be used, for example, to supply the air jets with an appropriate air pressure configured to divert/eject the material pieces 101 from the conveyor system 103.

Although the example illustrated in FIG. 1 uses air jets to divert/eject material pieces, other mechanisms may be used to divert/eject the material pieces, such as robotically removing the material pieces from the conveyor belt, pushing the material pieces from the conveyor belt (e.g., with paint brush type plungers), causing an opening (e.g., a trap door) in the conveyor system 103 from which a material piece may drop, or using air jets to separate the material pieces into separate bins as they fall from the edge of the conveyor belt. A pusher device, as that term is used herein, may refer to any form of device which may be activated to dynamically displace an object on or from a conveyor system/device, employing pneumatic, mechanical, or other means to do so, such as any appropriate type of mechanical pushing mechanism (e.g., an ACME screw drive), pneumatic pushing mechanism, or air jet pushing mechanism. Some embodiments may include multiple pusher devices located at different locations and/or with different diversion path orientations along the path of the conveyor system. In various different implementations, these sorting systems describe herein may determine which pusher device to activate (if any) depending on characteristics of material pieces identified by the machine learning system. Moreover, the determination of which pusher device to activate may be based on the detected presence and/or characteristics of other objects that may also be within the diversion path of a pusher device concurrently with a target item. Furthermore, even for facilities where singulation along the conveyor system is not perfect, the disclosed sorting systems can recognize when multiple objects are not well singulated, and dynamically select from a plurality of pusher devices which should be activated based on which pusher device provides the best diversion path for potentially separating objects within close proximity. In some embodiments, objects identified as target objects may represent material that should be diverted off of the conveyor system. In other embodiments, objects identified as target objects represent material that should be allowed to remain on the conveyor system so that non-target materials are instead diverted.

In addition to the N sorting bins 136 . . . 139 into which material pieces 101 are diverted/ejected, the system 100 may also include a receptacle or bin 140 that receives material pieces 101 not diverted/ejected from the conveyor system 103 into any of the aforementioned sorting bins 136 . . . 139. For example, a material piece 101 may not be diverted/ejected from the conveyor system 103 into one of the N sorting bins 136 . . . 139 when the classification of the material piece 101 is not determined (or simply because the sorting devices failed to adequately divert/eject a piece), or when the material piece 101 contains a contaminant detected by the vision system 110 and/or the sensor system 120. Thus, the bin 140 may serve as a default receptacle into which unclassified material pieces are dumped. Alternatively, the bin 140 may be used to receive one or more classifications of material pieces that have deliberately not been assigned to any of the N sorting bins 136 . . . 139. These such material pieces may then be further sorted in accordance with other characteristics and/or by another sorting system.

Depending upon the variety of classifications of material pieces desired, multiple classifications may be mapped to a single sorting device and associated sorting bin. In other words, there need not be a one-to-one correlation between classifications and sorting bins. For example, it may be desired by the user to sort certain classifications of materials into the same sorting bin. To accomplish this sort, when a material piece 101 is classified as falling into a predetermined grouping of classifications, the same sorting device may be activated to sort these into the same sorting bin. Such combination sorting may be applied to produce any desired combination of sorted material pieces. The mapping of classifications may be programmed by the user (e.g., using the sorting algorithm (e.g., see FIG. 5) operated by the computer system 107) to produce such desired combinations. Additionally, the classifications of material pieces are user-definable, and not limited to any particular known classifications of material pieces.

The conveyor system 103 may include a circular conveyor (not shown) so that unclassified material pieces are returned to the beginning of the system 100 and run through the system 100 again. Moreover, because the system 100 is able to specifically track each material piece 101 as it travels on the conveyor system 103, some sort of sorting device (e.g., the sorting device 129) may be implemented to direct/eject a material piece 101 that the system 100 has failed to classify after a predetermined number of cycles through the system 100 (or the material piece 101 is collected in bin 140).

Within certain embodiments of the present disclosure, the conveyor system 103 may be divided into multiple belts configured in series such as, for example, two belts, where a first belt conveys the material pieces past the vision system 110 and/or an implemented sensor system 120, and a second belt conveys the material pieces from the vision system 110 and/or an implemented sensor system 120 to the sorting devices. Moreover, such a second conveyor belt may be at a lower height than the first conveyor belt, such that the material pieces fall from the first belt onto the second belt.

Within certain embodiments of the present disclosure that implement a sensor system 120, the emitting source 121 may be located above the detection area (i.e., above the conveyor system 103); however, certain embodiments of the present disclosure may locate the emitting source 121 and/or detectors 124 in other positions that still produce acceptable sensed/detected physical characteristics.

With systems 100 implementing an XRF system for a sensor system 120, signals representing the detected XRF spectrum may be converted into a discrete energy histogram such as on a per-channel (i.e., element) basis, as further described herein. Such a conversion process may be implemented within the control system 123, or the computer system 107. Within certain embodiments of the present disclosure, such a control system 123 or computer system 107 may include a commercially available spectrum acquisition module, such as the commercially available Amptech MCA 5000 acquisition card and software programmed to operate the card. Such a spectrum acquisition module, or other software implemented within the system 100, may be configured to implement a plurality of channels for dispersing x-rays into a discrete energy spectrum (i.e., histogram) with such a plurality of energy levels, whereby each energy level corresponds to an element that the system 100 has been configured to detect. The system 100 may be configured so that there are sufficient channels corresponding to certain elements within the chemical periodic table, which are important for distinguishing between different materials. The energy counts for each energy level may be stored in a separate collection storage register. The computer system 107 then reads each collection register to determine the number of counts for each energy level during the collection interval, and build the energy histogram. As will be described in more detail herein, a sorting algorithm configured in accordance with certain embodiments of the present disclosure may then utilize this collected histogram of energy levels to classify at least certain ones of the material pieces 101 and/or assist the vision system 110 in classifying the material pieces 101.

In accordance with certain embodiments of the present disclosure that implement an XRF system as the sensor system 120, the source 121 may include an in-line x-ray fluorescence (“IL-XRF”) tube, such as further described within U.S. Pat. No. 10,207,296. Such an IL-XRF tube may include a separate x-ray source each dedicated for one or more streams (e.g., singulated) of conveyed material pieces. In such a case, the one or more detectors 124 may be implemented as XRF detectors to detect fluoresced x-rays from material pieces 101 within each of the singulated streams. Examples of such XRF detectors are further described within U.S. Pat. No. 10,207,296.

It should be appreciated that, although the systems and methods described herein are described primarily in relation to classifying material pieces in solid state, the disclosure is not so limited. The systems and methods described herein may be applied to classifying a material having any of a range of physical states, including, but not limited to a liquid, molten, gaseous, or powdered solid state, another state, and any suitable combination thereof.

The systems and methods described herein may be applied to classify and/or sort individual material pieces having any of a variety of sizes as small as a ¼ inch in diameter or less. Even though the systems and methods described herein are described primarily in relation to sorting individual material pieces of a singulated stream one at a time, the systems and methods described herein are not limited thereto. Such systems and methods may be used to stimulate and/or detect emissions from a plurality of materials concurrently. For example, as opposed to a singulated stream of materials being conveyed along one or more conveyor belts in series, multiple singulated streams may be conveyed in parallel. Each stream may be a on a same belt or on different belts arranged in parallel. Further, pieces may be randomly distributed on (e.g., across and along) one or more conveyor belts. Accordingly, the systems and methods described herein may be used to stimulate, and/or detect emissions from, a plurality of these small pieces at the same time. In other words, a plurality of small pieces may be treated as a single piece as opposed to each small piece being considered individually. Accordingly, the plurality of small pieces of material may be classified and sorted (e.g., diverted/ejected from the conveyor system) together. It should be appreciated that a plurality of larger material pieces also may be treated as a single material piece.

As previously noted, certain embodiments of the present disclosure may implement one or more vision systems (e.g., vision system 110) in order to identify, track, and/or classify material pieces. In accordance with embodiments of the present disclosure, such a vision system(s) may operate alone to identify and/or classify and sort material pieces, or may operate in combination with a sensor system (e.g., sensor system 120) to identify and/or classify and sort material pieces. If a sorting system (e.g., system 100) is configured to operate solely with such a vision system(s) 110, then the sensor system 120 may be omitted from the system 100 (or simply deactivated).

Such a vision system may be configured with one or more devices for capturing or acquiring images of the material pieces as they pass by on a conveyor system. The devices may be configured to capture or acquire any desired range of wavelengths irradiated or reflected by the material pieces, including, but not limited to, visible, infrared (“IR”), ultraviolet (“UV”) light. For example, the vision system may be configured with one or more cameras (still and/or video, either of which may be configured to capture two-dimensional, three-dimensional, and/or holographical images) positioned in proximity (e.g., above) the conveyor system so that images of the material pieces are captured as they pass by the sensor system(s). In accordance with alternative embodiments of the present disclosure, data captured by a sensor system 120 may be processed (converted) into data to be utilized (either solely or in combination with the image data captured by the vision system 110) for classifying/sorting of the material pieces. Such an implementation may be in lieu of, or in combination with, utilizing the sensor system 120 for classifying material pieces.

Regardless of the type(s) of sensed characteristics/information captured of the material pieces, the information may then be sent to a computer system (e.g., computer system 107) to be processed by a machine learning system in order to identify and/or classify each of the material pieces. Such a machine learning system may implement any well-known machine learning system, including one that implements a neural network (e.g., artificial neural network, deep neural network, convolutional neural network, recurrent neural network, autoencoders, reinforcement learning, etc.), supervised learning, unsupervised learning, semi-supervised learning, reinforcement learning, self learning, feature learning, sparse dictionary learning, anomaly detection, robot learning, association rule learning, fuzzy logic, artificial intelligence (“AI”), deep learning algorithms, deep structured learning hierarchical learning algorithms, support vector machine (“SVM”) (e.g., linear SVM, nonlinear SVM, SVM regression, etc.), decision tree learning (e.g., classification and regression tree (“CART”), ensemble methods (e.g., ensemble learning, Random Forests, Bagging and Pasting, Patches and Subspaces, Boosting, Stacking, etc.), dimensionality reduction (e.g., Projection, Manifold Learning, Principal Components Analysis, etc.) and/or deep machine learning algorithms, such as those described in and publicly available at the deeplearning.net website (including all software, publications, and hyperlinks to available software referenced within this website), which is hereby incorporated by reference herein. Non-limiting examples of publicly available machine learning software and libraries that could be utilized within embodiments of the present disclosure include Python, OpenCV, Inception, Theano, Torch, PyTorch, Pylearn2, Numpy, Blocks, TensorFlow, MXNet, Caffe, Lasagne, Keras, Chainer, Matlab Deep Learning, CNTK, MatConvNet (a MATLAB toolbox implementing convolutional neural networks for computer vision applications), DeepLearnToolbox (a Matlab toolbox for Deep Learning (from Rasmus Berg Palm)), BigDL, Cuda-Convnet (a fast C++/CUDA implementation of convolutional (or more generally, feed-forward) neural networks), Deep Belief Networks, RNNLM, RNNLIB-RNNLIB, matrbm, deeplearning4j, Eblearn.Ish, decpmat, MShadow, Matplotlib, SciPy, CXXNET, Nengo-Nengo, Eblearn, cudamat, Gnumpy, 3-way factored RBM and mcRBM, mPOT (Python code using CUDAMat and Gnumpy to train models of natural images), ConvNet, Elektronn, OpenNN, NeuralDesigner, Theano Generalized Hebbian Learning, Apache Singa, Lightnet, and SimpleDNN.

Machine learning often occurs in two stages. For example, first, training occurs, which may be performed offline in that the system 100 is not being utilized to perform actual classifying/sorting of material pieces. The system 100 may be utilized to train the machine learning system in that homogenous sets (also referred to herein as control samples) of material pieces (i.e., having the same types or classes of materials) are passed through the system 100 (e.g., by a conveyor system 103); and all such material pieces may not be sorted, but may be collected in a common bin (e.g., bin 140). Alternatively, the training may be performed at another location remote from the system 100, including using some other mechanism for collecting sensed information (characteristics) of homogenous sets of material pieces. During this training stage, algorithms within the machine learning system extract features from the captured information (e.g., using image processing techniques well known in the art). Non-limiting examples of training algorithms include, but are not limited to, linear regression, gradient descent, feed forward, polynomial regression, learning curves, regularized learning models, and logistic regression. It is during this training stage that the algorithms within the machine learning system learn the relationships between different types of materials and their features/characteristics (e.g., as captured by the vision system and/or sensor system(s)), creating a knowledge base for later classification of a heterogeneous mixture of material pieces received by the system 100 for sorting by desired classifications. Such a knowledge base may include one or more libraries, wherein each library includes parameters (also referred to herein as “neural network parameters”) for utilization by the machine learning system in classifying material pieces. For example, one particular library may include parameters configured by the training stage to recognize and classify a particular type or class of material. In accordance with certain embodiments of the present disclosure, such libraries may be inputted into the machine learning system and then the user of the system 100 may be able to adjust certain ones of the parameters in order to adjust an operation of the system 100 (for example, adjusting the threshold effectiveness of how well the machine learning system recognizes a particular material from a heterogeneous mixture of materials).

Additionally, the inclusion of certain materials (e.g., chemical elements or compounds) in material pieces (e.g., metal alloys), or combinations of certain chemical elements or compounds, result in identifiable physical features (e.g., visually discernible characteristics) in materials. As a result, when a plurality of material pieces containing such a particular composition are passed through the aforementioned training stage, the machine learning system can learn how to distinguish such material pieces from others. Consequently, a machine learning system configured in accordance with certain embodiments of the present disclosure may be configured to sort between material pieces as a function of their respective material/chemical compositions. For example, such a machine learning system may be configured so that aluminum alloys can be sorted as a function of the percentage of a specified alloying material contained within the aluminum alloys.

For example, FIG. 2 shows captured or acquired images of exemplary material pieces of cast aluminum, which may be used during the aforementioned training stage. FIG. 3 shows captured or acquired images of exemplary material pieces of extruded aluminum, which may be used during the aforementioned training stage. FIG. 4 shows captured or acquired images of exemplary material pieces of wrought aluminum, which may be used during the aforementioned training stage. During the training stage, a plurality of material pieces of a particular (homogenous) classification (type) of material, which are the control samples, may be delivered past the vision system by the conveyor system so that the machine learning system detects, extracts, and learns what features visually represent such exemplary material pieces. In other words, images of cast aluminum material pieces such as shown in FIG. 2 may be first passed through such a training stage so that the machine learning algorithm “learns” how to detect, recognize, and classify material pieces composed of cast aluminum alloys. This creates a library of parameters particular to cast aluminum material pieces. Then, the same process can be performed with respect to images of extruded aluminum material pieces, such as shown in FIG. 3, creating a library of parameters particular to extruded aluminum material pieces. And, the same process can be performed with respect to images of wrought aluminum material pieces, such as shown in FIG. 4, creating a library of parameters particular to wrought aluminum material pieces. For each type of material to be classified by the vision system, any number of exemplary material pieces of that type of material may be passed by the vision system. Given a captured image as input data, the machine learning algorithms may use N classifiers, each of which test for one of N different material types.

After the algorithms have been established and the machine learning system has sufficiently learned the differences for the material classifications (e.g., within a user-defined level of statistical confidence), the libraries of neural network parameters for the different materials are then implemented into a material classifying and/or sorting system (e.g., system 100) to be used for identifying and/or classifying material pieces from a heterogeneous mixture of material pieces, and then possibly sorting such classified material pieces if sorting is to be performed.

Techniques to construct, optimize, and utilize a machine learning system are known to those of ordinary skill in the art as found in relevant literature. Examples of such literature include the publications: Krizhevsky et al., “ImageNet Classification with Deep Convolutional Networks,” Proceedings of the 25th International Conference on Neural Information Processing Systems, Dec. 3-6, 2012, Lake Tahoe, Nev., and LeCun et al., “Gradient-Based Learning Applied to Document Recognition,” Proceedings of the IEEE, Institute of Electrical and Electronic Engineers (IEEE), November 1998, both of which are hereby incorporated by reference herein in their entirety.

In an example technique, data captured by a sensor and/or vision system with respect to a particular material piece may be processed as an array of data values. For example, the data may be image data captured by a digital camera or other type of imaging sensor with respect to a particular material piece and processed as an array of pixel values. Each data value may be represented by a single number, or as a series of numbers representing values. These values are multiplied by the neuron weight parameters, and may possibly have a bias added. This is fed into a neuron nonlinearity. The resulting number output by the neuron can be treated much as the values were, with this output multiplied by subsequent neuron weight values, a bias optionally added, and once again fed into a neuron nonlinearity. Each such iteration of the process is known as a “layer” of the neural network. The final outputs of the final layer may be interpreted as probabilities that a material is present or absent in the captured data pertaining to the material piece. Examples of such a process are described in detail in both of the previously noted “ImageNet Classification with Deep Convolutional Networks” and “Gradient-Based Learning Applied to Document Recognition” references.

In accordance with embodiments of the present disclosure, as a final layer (the “classification layer”), the final set of neurons' outputs is trained to represent the likelihood a material piece is associated with the captured data. During operation, if the likelihood that a material piece is associated with the captured data is over a user-specified threshold, then it is determined that the particular material piece is indeed associated with the captured data. These techniques can be extended to determine not only the presence of a type of material associated with particular captured data, but also whether sub-regions of the particular captured data belong to one type of material or another type of material. This process is known as segmentation, and techniques to use neural networks exist in the literature, such as those known as “fully convolutional” neural networks, or networks that otherwise include a convolutional portion (i.e., are partially convolutional), if not fully convolutional. This allows for material location and size to be determined.

It should be understood that the present disclosure is not exclusively limited to machine learning techniques. Other common techniques for material classification/identification may also be used. For instance, a sensor system may utilize optical spectrometric techniques using multi- or hyper-spectral cameras to provide a signal that may indicate the presence or absence of a type of material by examining the spectral emissions of the material. Photographs of a material piece may also be used in a template-matching algorithm, wherein a database of images is compared against an acquired image to find the presence or absence of certain types of materials from that database. A histogram of the captured image may also be compared against a database of histograms. Similarly, a bag of words model may be used with a feature extraction technique, such as scale-invariant feature transform (“SIFT”), to compare extracted features between a captured image and those in a database.

Therefore, as disclosed herein, certain embodiments of the present disclosure provide for the identification/classification of one or more different materials in order to determine which material pieces should be diverted from a conveyor system or device. In accordance with certain embodiments, machine learning techniques are utilized to train (i.e., configure) a neural network to identify a variety of one or more different materials. Images, or other types of sensed information, are captured of materials (e.g., traveling on a conveyor system), and based on the identification/classification of such materials, the systems described herein can decide which material piece should be allowed to remain on the conveyor system, and which should be diverted/removed from the conveyor system (for example, either into a collection bin, or diverted onto another conveyor system).

In accordance with certain embodiments of the present disclosure, a machine learning system for an existing installation may be dynamically reconfigured to detect and recognize characteristics of a new material by replacing a current set of neural network parameters with a new set of neural network parameters.

One point of mention here is that, in accordance with certain embodiments of the present disclosure, the detected/extracted features/characteristics of the material pieces may not be necessarily simply particularly identifiable physical characteristics; they can be abstract formulations that can only be expressed mathematically, or not mathematically at all; nevertheless, the machine learning system parses all of the data to look for patterns that allow the control samples to be classified during the training stage. Furthermore, the machine learning system may take subsections of captured information of a material piece and attempt to find correlations between the pre-defined classifications.

In accordance with certain embodiments of the present disclosure, instead of utilizing a training stage whereby control samples of material pieces are passed by the vision system and/or sensor system(s), training of the machine learning system may be performed utilizing a labeling/annotation technique (or any other supervised learning technique) whereby as data/information of material pieces are captured by a vision/sensor system, a user inputs a label or annotation that identifies each material piece, which is then used to create the library for use by the machine learning system when classifying material pieces within a heterogenous mixture of material pieces.

In accordance with certain embodiments of the present disclosure, any sensed characteristics output by any of the sensor systems 120 disclosed herein may be input into a machine learning system in order to classify and/or sort materials. For example, in a machine learning system implementing supervised learning, sensor system 120 outputs that uniquely characterize a particular type or composition of material (e.g., a particular metal alloy) may be used to train the machine learning system.

After going through a shredder, sidings (typically made from thin aluminum sheets), extrusions (typically manufactured from thick aluminum framing bars), and castings look very different. FIG. 2 shows visual images of exemplary scrap pieces from cast aluminum. FIG. 3 shows visual images of exemplary scrap pieces from aluminum extrusions. FIG. 4 shows visual images of exemplary scrap pieces from wrought aluminum. Embodiments of the present disclosure utilize a vision system as described herein capable of classifying/sorting between these three different types of aluminum scrap pieces. As shown by the examples in FIGS. 2-4, aluminum extrusions have an overall physical appearance that is distinguishable from cast and wrought aluminum scrap pieces, which can be learned by a machine learning system configured in accordance with embodiments of the present disclosure.

Embodiments of the present disclosure are configured to sort the wrought aluminum alloy material pieces from the Twitch, which contains both wrought and cast aluminum pieces. In certain embodiments of the present disclosure, extruded aluminum alloy pieces can be sorted with the wrought aluminum alloy pieces (or sorted separately from both cast and wrought aluminum). Since most of the Mg is within the wrought aluminum, the remaining aluminum scrap pieces, containing mostly cast aluminum alloys, have relatively insignificant amounts of Mg. In accordance with certain embodiments of the present disclosure, another sort (or plurality of sorting cycles) can be performed on these remaining aluminum scrap pieces (also referred to herein as the “cast fraction”) in order to classify/sort between any plurality of different cast aluminum alloys and/or to remove other impurities (e.g., scrap pieces composed of PCB, stainless steel, foam, rubber, etc.). The cast fraction may include cast alloys such as 319, 356, 360, and/or 380 series alloy pieces. These alloys contain varying amounts of silicon, Cu, Zn, Fe, and Mn, but contain extremely small amounts of Mg, typically 0-0.6%.

In accordance with certain embodiments of the present disclosure, one or more of the sensor systems 120 disclosed herein may be utilized to classify/sort either or both of the aforementioned cast fractions and wrought fractions. For example, one or both of an XRF system and/or a sensor system using LIBS may be utilized to classify/sort between two or more different cast aluminum alloys or two or more different wrought aluminum alloys. The utilization of an XRF system to do so is disclosed in U.S. Pat. No. 10,207,296.

The spectroscopy technique known as Laser-Induced Breakdown Spectroscopy (“LIBS), Laser Spark Spectroscopy (“LSS”), or Laser-Induced Optical Emission Spectroscopy (“LIOES”) uses a focused laser beam to vaporize and subsequently produce spectral line emissions from a sample material. In this way samples placed at a distance from the analyzing instrumentation, can be analyzed for their chemical composition. Embodiments of the present disclosure can utilize any one of the foregoing to classify a plurality of materials into different classes for sorting. The use of LIBS for sorting is further described in U.S. Pat. Nos. 5,042,947, 6,545,240, and 10,478,861, all of which are hereby incorporated by reference herein.

FIG. 5 illustrates a flowchart diagram depicting exemplary embodiments of a process 3500 of classifying/sorting material pieces utilizing a vision system and/or sensor system in accordance with certain embodiments of the present disclosure. The process 3500 may be configured to operate within any of the embodiments of the present disclosure described herein, including the system 100 of FIG. 1. Operation of the process 3500 may be performed by hardware and/or software, including within a computer system (e.g., computer system 3400 of FIG. 8) controlling the sorting system (e.g., the computer system 107, the vision system 110, and/or the sensor system(s) 120 of FIG. 1). In the process block 3501, the material pieces are fed onto a conveyor system. In the process block 3502, the location on the conveyor system of each material piece is detected for tracking of each material piece as it travels through the sorting system. This may be performed by the vision system 110 (for example, by distinguishing a material piece from the underlying conveyor system material while in communication with a conveyor system position detector (e.g., the position detector 105)). Alternatively, a linear sheet laser beam can be used to locate the pieces. Or, any system that can create a light source (including, but not limited to, visual light, UV, and IR) and have a detector that can be used to locate the pieces. In the process block 3503, when a material piece has traveled in proximity to one or more of the vision system and/or the sensor system(s), sensed information/characteristics of the material piece is captured/acquired. In the process block 3504, a vision system (e.g., implemented within the computer system 107), such as previously disclosed, may perform pre-processing of the captured information, which may be utilized to detect (extract) each of the material pieces (e.g., from the background (e.g., the conveyor belt); in other words, the pre-processing may be utilized to identify the difference between the material piece and the background). Well-known image processing techniques such as dilation, thresholding, and contouring may be utilized to identify the material piece as being distinct from the background. In the process block 3505, segmentation may be performed. For example, the captured information may include information pertaining to one or more material pieces. Additionally, a particular material piece may be located on a seam of the conveyor belt when its image is captured. Therefore, it may be desired in such instances to isolate the image of an individual material piece from the background of the image. In an exemplary technique for the process block 3505, a first step is to apply a high contrast of the image; in this fashion, background pixels are reduced to substantially all black pixels, and at least some of the pixels pertaining to the material piece are brightened to substantially all white pixels. The image pixels of the material piece that are white are then dilated to cover the entire size of the material piece. After this step, the location of the material piece is a high contrast image of all white pixels on a black background. Then, a contouring algorithm can be utilized to detect boundaries of the material piece. The boundary information is saved, and the boundary locations are then transferred to the original image. Segmentation is then performed on the original image on an area greater than the boundary that was earlier defined. In this fashion, the material piece is identified and separated from the background.

In the optional process block 3506, the material pieces may be conveyed along the conveyor system within proximity of a distance measuring device and/or a sensor system in order to determine a size and/or shape of the material pieces, which may be useful if an XRF system, LIBS system, or some other spectroscopy sensor is also implemented within the sorting system and requires such size and/or shape determinations. In the process block 3507, post processing may be performed. Post processing may involve resizing the captured information/data to prepare it for use in the neural networks. This may also include modifying certain properties (e.g., enhancing image contrast, changing the image background, or applying filters) in a manner that will yield an enhancement to the capability of the machine learning system to classify the material pieces. In the process block 3509, the data may be resized. Data resizing may be necessary under certain circumstances to match the data input requirements for certain machine learning systems, such as neural networks. For example, neural networks may require much smaller image sizes (e.g., 225×255 pixels or 299×299 pixels) than the sizes of the images captured by typical digital cameras. Moreover, the smaller the input data size, the less processing time is needed to perform the classification. Thus, smaller data sizes can ultimately increase the throughput of the sorter system 100 and increase its value.

In the process blocks 3510 and 3511, for each material piece, the type or class of material is identified/classified based on the sensed/detected features. For example, the process block 3510 may be configured with a neural network employing one or more machine learning algorithms, which compare the extracted features with those stored in the knowledge base generated during the training stage, and assigns the classification with the highest match to each of the material pieces based on such a comparison. The algorithms of the machine learning system may process the captured information/data in a hierarchical manner by using automatically trained filters. The filter responses are then successfully combined in the next levels of the algorithms until a probability is obtained in the final step. In the process block 3511, these probabilities may be used for each of the N classifications to decide into which of the N sorting bins the respective material pieces should be sorted. For example, each of the N classifications may be assigned to one sorting bin, and the material piece under consideration is sorted into that bin that corresponds to the classification returning the highest probability larger than a predefined threshold. Within embodiments of the present disclosure, such predefined thresholds may be preset by the user. A particular material piece may be sorted into an outlier bin (e.g., sorting bin 140) if none of the probabilities is larger than the predetermined threshold.

Next, in the process block 3512, a sorting device corresponding to the classification, or classifications, of the material piece may be activated. Between the time at which the image of the material piece was captured and the time at which the sorting device is activated, the material piece has moved from the proximity of the vision system and/or sensor system(s) to a location downstream on the conveyor system (e.g., at the rate of conveying of a conveyor system). In embodiments of the present disclosure, the activation of the sorting device is timed such that as the material piece passes the sorting device mapped to the classification of the material piece, the sorting device is activated, and the material piece is diverted/ejected from the conveyor system into its associated sorting bin. Within embodiments of the present disclosure, the activation of a sorting device may be timed by a respective position detector that detects when a material piece is passing before the sorting device and sends a signal to enable the activation of the sorting device. In the process block 3513, the sorting bin corresponding to the sorting device that was activated receives the diverted/ejected material piece.

FIG. 6 illustrates a flowchart diagram depicting exemplary embodiments of a process 400 of sorting material pieces in accordance with certain embodiments of the present disclosure. The process 400 may be configured to operate within any of the embodiments of the present disclosure described herein, including the system 100 of FIG. 1. The process 400 may be configured to operate in conjunction with the process 3500. For example, in accordance with certain embodiments of the present disclosure, the process blocks 403 and 404 may be incorporated in the process 3500 (e.g., operating in series or in parallel with the process blocks 3503-3510) in order to combine the efforts of a vision system 110 that is implemented in conjunction with a machine learning system with a sensor system (e.g., the sensor system 120) that is not implemented in conjunction with a machine learning system in order to classify and/or sort material pieces.

Operation of the process 400 may be performed by hardware and/or software, including within a computer system (e.g., computer system 3400 of FIG. 8) controlling the sorting system (e.g., the computer system 107 of FIG. 1). In the process block 401, the material pieces are fed onto a conveyor system. Next, in the optional process block 402, the material pieces may be conveyed along the conveyor system within proximity of a distance measuring device and/or an optical imaging system in order to determine a size and/or shape of the material pieces. In the process block 403, when a material piece has traveled in proximity of the sensor system, the material piece may be interrogated, or stimulated, with some type of energy appropriate for the particular type of sensor technology utilized by the sensor system (e.g., a LIBS system). In the process block 404, physical characteristics of the material piece are sensed/detected by the sensor system. In the process block 405, for at least some of the material pieces, the type of material is identified/classified based (at least in part) on the sensed/detected characteristics, which may be combined with the classification by the machine learning system in conjunction with the vision system 110.

Next, if sorting of the material pieces is to be performed, in the process block 406, a sorting device corresponding to the classification, or classifications, of the material piece is activated. Between the time at which the material piece was sensed and the time at which the sorting device is activated, the material piece has moved from the proximity of the sensor system to a location downstream on the conveyor system, at the rate of conveying of the conveyor system. In certain embodiments of the present disclosure, the activation of the sorting device is timed such that as the material piece passes the sorting device mapped to the classification of the material piece, the sorting device is activated, and the material piece is diverted/ejected from the conveyor system into its associated sorting bin. Within certain embodiments of the present disclosure, the activation of a sorting device may be timed by a respective position detector that detects when a material piece is passing before the sorting device and sends a signal to enable the activation of the sorting device. In the process block 407, the sorting bin corresponding to the sorting device that was activated receives the diverted/ejected material piece.

In accordance with certain embodiments of the present disclosure, a plurality of at least a portion of the system 100 may be linked together in succession in order to perform multiple iterations or layers of sorting. For example, when two or more systems 100 are linked in such a manner, the conveyor system may be implemented with a single conveyor belt, or multiple conveyor belts, conveying the material pieces past a first vision system (and, in accordance with certain embodiments, a sensor system) configured for sorting material pieces of a first set of a heterogeneous mixture of materials by a sorter (e.g., the first automation control system 108 and associated one or more sorting devices 126 . . . 129) into a first set of one or more receptacles (e.g., sorting bins 136 . . . 139), and then conveying the material pieces past a second vision system (and, in accordance with certain embodiments, another sensor system) configured for sorting material pieces of a second set of a heterogeneous mixture of materials by a second sorter into a second set of one or more sorting bins.

Such successions of systems 100 can contain any number of such systems linked together in such a manner. In accordance with certain embodiments of the present disclosure, each successive vision system may be configured to sort out a different material than previous vision system(s).

In accordance with various embodiments of the present disclosure, different types or classes of materials may be classified by different types of sensors each for use with a machine learning system, and combined to classify material pieces in a stream of scrap or waste.

In accordance with various embodiments of the present disclosure, data from two or more sensors can be combined using a single or multiple machine learning systems to perform classifications of material pieces.

In accordance with various embodiments of the present disclosure, multiple sensor systems can be mounted onto a single conveyor system, with each sensor system utilizing a different machine learning system. In accordance with various embodiments of the present disclosure, multiple sensor systems can be mounted onto different conveyor systems, with each sensor system utilizing a different machine learning system.

FIGS. 7A-7B illustrate a system and process 1600 configured in accordance with certain embodiments of the present disclosure in order to sort a plurality of metal alloy pieces. FIG. 7A illustrates an exemplary non-limiting schematic diagram of a side view of such a system and process 1600, while FIG. 7B illustrates a top view. Though FIGS. 7A-7B depict three stages of classification/sorting, any number of such stages may be implemented in accordance with various embodiments of the present disclosure.

A plurality of metal alloy pieces 1601 may be conveyed (e.g., by a conveyor belt 1602) to be picked up by an inclined conveyor system 1603. Note that the material pieces 1601 are not depicted in FIG. 7B for the sake of simplicity. The conveyor system 1603 conveys the material pieces 1601 past a sensor system 1610 in order to classify the material pieces for sorting. Any of the disclosed vision system 110 or sensor systems 120 (e.g., LIBS, XRF, etc.) may be utilized.

In a non-limiting example, the material pieces 1601 fed onto the conveyor system 1602 may be a mixture of aluminum alloys that include cast, wrought, and/or extruded aluminum alloys of various alloy compositions. An AI system 1610 may be configured to recognize, classify, and distinguish those material pieces composed of wrought aluminum alloy(s) from those composed of cast aluminum alloys. The conveyor system 1603 may be configured to operate at a sufficient speed in order to “throw” the material pieces classified as wrought aluminum alloy(s) onto a following inclined conveyor system 1604. Material pieces not classified as composed of wrought aluminum alloy(s) (e.g., cast and/or extruded alloys) are ejected by a sorting device 1620 onto a lower positioned conveyor system 1606. For example, such a sorting device 1620 may be an air jet nozzle such as described herein, which is actuated to eject a material piece not classified as wrought aluminum alloy(s) from the normal trajectory of material pieces being “thrown” from the end of the conveyor system 1603 onto the conveyor system 1604. The material pieces not classified as wrought aluminum alloy(s) (e.g., cast and/or extruded alloys) may be conveyed into a bin or receptacle 1630, or they may be conveyed past another sensor system 120 as disclosed herein.

The material pieces classified as wrought aluminum alloy(s) may be conveyed past an XRF or LIBS system 1611, which may be configured to identify, classify, and distinguish between different wrought aluminum alloy(s), including with a same wrought aluminum alloy series. The conveyor system 1604 may be configured to operate at a sufficient speed in order to “throw” the material pieces classified as belonging to one or more specific wrought aluminum alloys onto a following inclined conveyor system 1605. The other wrought aluminum alloy(s) may be ejected by a sorting device 1621 onto a lower positioned conveyor system 1607. For example, such a sorting device 1621 may be an air jet nozzle such as described herein, which is actuated to eject a material piece classified as belonging to one or more specific wrought aluminum alloy(s) from the normal trajectory of material pieces being “thrown” from the end of the conveyor system 1604 onto the conveyor system 1605. The classified material pieces may be conveyed into a bin or receptacle 1631.

The material pieces classified as belonging to the one or more specific wrought aluminum alloy(s) may be conveyed past a sensor system 1612, which may be configured to identify and classify those material pieces that contain a threshold amount of a specific material in order to classify a specific wrought aluminum alloy that is known to contain such a specific material.

In accordance with alternative embodiments of the present disclosure, the cast aluminum alloy(s) previously sorted out by the sorter 1620 may be conveyed by the conveyor system 1606 past an XRF system as described herein in order to classify/sort out certain specific cast alloy fractions. Cast aluminum alloy 319 has a single large copper peak observable in its XRF spectrum, cast aluminum alloy 356 does not have such a large copper peak, and cast aluminum alloy 380 has both large copper and zinc peaks. These large differences can be utilized by an XRF system to sort between these cast aluminum alloys with high accuracy. Classifying/sorting of cast fractions is further disclosed in U.S. Published Patent Application No. 2021/0229133, which is hereby incorporated by reference herein.

The conveyor systems 1605 and 1608 may be configured to operate in a similar manner as the conveyor systems 1603 and 1604, the sorter 1622 may be configured to operate in a similar manner as the sorters 1620, 1621, and the bins 1632, 1633 may be configured similarly as the bins 1630, 1631.

Note that the system and process 1600 is not limited to one line of conveyor systems, but may be expanded to multiple lines each ejecting classified material pieces onto multiple conveyor systems (e.g., conveyor systems 1606 . . . 1608). Likewise, one or more of the conveyor systems 1606 . . . 1608 may be implemented with any number of additional sensor systems to further classify those material pieces.

Furthermore, embodiments of the present disclosure are not limited to the sorting of aluminum alloys, but may be configured to sort any number of different classes of materials, including, but not limited to, the sorting of various metals (e.g., copper, brass, zinc, aluminum, etc.) from Zorba.

With reference now to FIG. 8, a block diagram illustrating a data processing (“computer”) system 3400 is depicted in which aspects of embodiments of the disclosure may be implemented. (The terms “computer,” “system,” “computer system,” and “data processing system” may be used interchangeably herein.) The computer system 107, the automation control system 108, aspects of the sensor system(s) 120, and/or the vision system 110 may be configured similarly as the computer system 3400. The computer system 3400 may employ a local bus 3405 (e.g., a peripheral component interconnect (“PCI”) local bus architecture). Any suitable bus architecture may be utilized such as Accelerated Graphics Port (“AGP”) and Industry Standard Architecture (“ISA”), among others. One or more processors 3415, volatile memory 3420, and non-volatile memory 3435 may be connected to the local bus 3405 (e.g., through a PCI Bridge (not shown)). An integrated memory controller and cache memory may be coupled to the one or more processors 3415. The one or more processors 3415 may include one or more central processor units and/or one or more graphics processor units and/or one or more tensor processing units. Additional connections to the local bus 3405 may be made through direct component interconnection or through add-in boards. In the depicted example, a communication (e.g., network (LAN)) adapter 3425, an I/O (e.g., small computer system interface (“SCSI”) host bus) adapter 3430, and expansion bus interface (not shown) may be connected to the local bus 3405 by direct component connection. An audio adapter (not shown), a graphics adapter (not shown), and display adapter 3416 (coupled to a display 3440) may be connected to the local bus 3405 (e.g., by add-in boards inserted into expansion slots).

The user interface adapter 3412 may provide a connection for a keyboard 3413 and a mouse 3414, modem (not shown), and additional memory (not shown). The I/O adapter 3430 may provide a connection for a hard disk drive 3431, a tape drive 3432, and a CD-ROM drive (not shown).

An operating system may be run on the one or more processors 3415 and used to coordinate and provide control of various components within the computer system 3400. In FIG. 8, the operating system may be a commercially available operating system. An object-oriented programming system (e.g., Java, Python, etc.) may run in conjunction with the operating system and provide calls to the operating system from programs or programs (e.g., Java, Python, etc.) executing on the system 3400. Instructions for the operating system, the object-oriented operating system, and programs may be located on non-volatile memory 3435 storage devices, such as a hard disk drive 3431, and may be loaded into volatile memory 3420 for execution by the processor 3415.

Those of ordinary skill in the art will appreciate that the hardware in FIG. 8 may vary depending on the implementation. Other internal hardware or peripheral devices, such as flash ROM (or equivalent nonvolatile memory) or optical disk drives and the like, may be used in addition to or in place of the hardware depicted in FIG. 8. Also, any of the processes of the present disclosure may be applied to a multiprocessor computer system, or performed by a plurality of such systems 3400. For example, training of the vision system 110 may be performed by a first computer system 3400, while operation of the vision system 110 for sorting may be performed by a second computer system 3400.

As another example, the computer system 3400 may be a stand-alone system configured to be bootable without relying on some type of network communication interface, whether or not the computer system 3400 includes some type of network communication interface. As a further example, the computer system 3400 may be an embedded controller, which is configured with ROM and/or flash ROM providing non-volatile memory storing operating system files or user-generated data.

The depicted example in FIG. 8 and above-described examples are not meant to imply architectural limitations. Further, a computer program form of aspects of the present disclosure may reside on any computer readable storage medium (i.e., floppy disk, compact disk, hard disk, tape, ROM, RAM, etc.) used by a computer system.

As has been described herein, embodiments of the present disclosure may be implemented to perform the various functions described for identifying, tracking, classifying, and/or sorting material pieces. Such functionalities may be implemented within hardware and/or software, such as within one or more data processing systems (e.g., the data processing system 3400 of FIG. 8), such as the previously noted computer system 107, the vision system 110, aspects of the sensor system(s) 120, and/or the automation control system 108. Nevertheless, the functionalities described herein are not to be limited for implementation into any particular hardware/software platform.

As will be appreciated by one skilled in the art, aspects of the present disclosure may be embodied as a system, process, method, and/or program product. Accordingly, various aspects of the present disclosure may take the form of an entirely hardware embodiment, an entirely software embodiment (including firmware, resident software, micro-code, etc.), or embodiments combining software and hardware aspects, which may generally be referred to herein as a “circuit,” “circuitry,” “module,” or “system.” Furthermore, aspects of the present disclosure may take the form of a program product embodied in one or more computer readable storage medium(s) having computer readable program code embodied thereon. (However, any combination of one or more computer readable medium(s) may be utilized. The computer readable medium may be a computer readable signal medium or a computer readable storage medium.)

A computer readable storage medium may be, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, biologic, atomic, or semiconductor system, apparatus, controller, or device, or any suitable combination of the foregoing, wherein the computer readable storage medium is not a transitory signal per se. More specific examples (a non-exhaustive list) of the computer readable storage medium may include the following: an electrical connection having one or more wires, a portable computer diskette, a hard disk, a random access memory (“RAM”) (e.g., RAM 3420 of FIG. 8), a read-only memory (“ROM”) (e.g., ROM 3435 of FIG. 8), an erasable programmable read-only memory (“EPROM” or flash memory), an optical fiber, a portable compact disc read-only memory (“CD-ROM”), an optical storage device, a magnetic storage device (e.g., hard drive 3431 of FIG. 8), or any suitable combination of the foregoing. In the context of this document, a computer readable storage medium may be any tangible medium that can contain or store a program for use by or in connection with an instruction execution system, apparatus, controller, or device. Program code embodied on a computer readable signal medium may be transmitted using any appropriate medium, including but not limited to wireless, wire line, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

A computer readable signal medium may include a propagated data signal with computer readable program code embodied therein, for example, in baseband or as part of a carrier wave. Such a propagated signal may take any of a variety of forms, including, but not limited to, electromagnetic, optical, or any suitable combination thereof. A computer readable signal medium may be any computer readable medium that is not a computer readable storage medium and that can communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, controller, or device.

The flowchart and block diagrams in the figures illustrate architecture, functionality, and operation of possible implementations of systems, methods, processes, and program products according to various embodiments of the present disclosure. In this regard, each block in the flowcharts or block diagrams may represent a module, segment, or portion of code, which includes one or more executable program instructions for implementing the specified logical function(s). It should also be noted that, in some implementations, the functions noted in the blocks may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved.

Modules implemented in software for execution by various types of processors (e.g., GPU 3401, CPU 3415) may, for instance, include one or more physical or logical blocks of computer instructions, which may, for instance, be organized as an object, procedure, or function. Nevertheless, the executables of an identified module need not be physically located together, but may include disparate instructions stored in different locations which, when joined logically together, include the module and achieve the stated purpose for the module. Indeed, a module of executable code may be a single instruction, or many instructions, and may even be distributed over several different code segments, among different programs, and across several memory devices. Similarly, operational data (e.g., material classification libraries described herein) may be identified and illustrated herein within modules, and may be embodied in any suitable form and organized within any suitable type of data structure. The operational data may be collected as a single data set, or may be distributed over different locations including over different storage devices. The data may provide electronic signals on a system or network.

These program instructions may be provided to one or more processors and/or controller(s) of a general purpose computer, special purpose computer, or other programmable data processing apparatus (e.g., controller) to produce a machine, such that the instructions, which execute via the processor(s) (e.g., GPU 3401, CPU 3415) of the computer or other programmable data processing apparatus, create circuitry or means for implementing the functions/acts specified in the flowchart and/or block diagram block or blocks.

It will also be noted that each block of the block diagrams and/or flowchart illustrations, and combinations of blocks in the block diagrams and/or flowchart illustrations, can be implemented by special purpose hardware-based systems (e.g., which may include one or more graphics processing units (e.g., GPU 3401)) that perform the specified functions or acts, or combinations of special purpose hardware and computer instructions. For example, a module may be implemented as a hardware circuit including custom VLSI circuits or gate arrays, off-the-shelf semiconductors such as logic chips, transistors, controllers, or other discrete components. A module may also be implemented in programmable hardware devices such as field programmable gate arrays, programmable array logic, programmable logic devices, or the like.

Computer program code, i.e., instructions, for carrying out operations for aspects of the present disclosure may be written in any combination of one or more programming languages, including an object oriented programming language such as Java, Smalltalk, Python, C++, or the like, conventional procedural programming languages, such as the “C” programming language or similar programming languages, programming languages such as MATLAB or LabVIEW, or any of the machine learning software disclosed herein. The program code may execute entirely on the user's computer system, partly on the user's computer system, as a stand-alone software package, partly on the user's computer system (e.g., the computer system utilized for sorting) and partly on a remote computer system (e.g., the computer system utilized to train the machine learning system), or entirely on the remote computer system or server. In the latter scenario, the remote computer system may be connected to the user's computer system through any type of network, including a local area network (“LAN”) or a wide area network (“WAN”), or the connection may be made to an external computer system (for example, through the Internet using an Internet Service Provider). As an example of the foregoing, various aspects of the present disclosure may be configured to execute on one or more of the computer system 107, automation control system 108, the vision system 110, and aspects of the sensor system(s) 120.

These program instructions may also be stored in a computer readable storage medium that can direct a computer system, other programmable data processing apparatus, controller, or other devices to function in a particular manner, such that the instructions stored in the computer readable medium produce an article of manufacture including instructions which implement the function/act specified in the flowchart and/or block diagram block or blocks.

The program instructions may also be loaded onto a computer, other programmable data processing apparatus, controller, or other devices to cause a series of operational steps to be performed on the computer, other programmable apparatus or other devices to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide processes for implementing the functions/acts specified in the flowchart and/or block diagram block or blocks.