Predictive Process Control for Manufacturing Processes
Through the deep learning controller real-time analysis and adjustment of control values in the manufacturing process, the shortcomings of traditional manufacturing process controllers in terms of dynamic adjustment and cross-site trend considerations are solved, and the optimization of the manufacturing process and the specification compliance of the final output are achieved, reducing the quality of the production process and the reduction of the quality of the production process. output variability.
Patent Information
- Application Number
- CN202080044987.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Priority Date
- 2019-07-23
- Filing Date
- 2020-06-23
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2040-06-23
AI Technical Summary
Traditional manufacturing process controllers have limited considerations in terms of dynamic adjustment and cross-site trends, making it difficult to ensure the optimization of the manufacturing process and the specification compliance of the final output, and there is high variability.
Based on the machine learning model, deep learning controllers are adopted to receive and analyze control values from multiple machining stations in real time, predict specifications of intermediate outputs and final outputs, dynamically adjust control inputs to ensure compliance with specifications, and optimize the manufacturing process by identifying key influencing factors.
It realizes dynamic adjustment of control input without interrupting the manufacturing process to ensure that the final output meets specifications, optimizes the manufacturing process, reduces output variability, and improves production stability and efficiency.
Smart Images

Figure CN113994286B_ABST
Abstract
Description
[0001] CROSS-REFERENCE TO RELATED APPLICATIONS
[0002] This application claims priority to U.S. patent application No. 16 / 519,102, filed on July 23, 2019, entitled “PREDICTIVE PROCESS CONTROL FOR A MANUFACTURING PROCESS,” which claims priority to U.S. application No. 62 / 865,859, filed on June 24, 2019, entitled “SYSTEMS, APPARATUS AND METHODS FOR PREDICTIVE PROCESS CONTROL OF THE MANUFACTURING PROCESS,” the entire contents of which are incorporated herein by reference in their entirety. Technical Field
[0003] The present disclosure generally relates to systems, devices and methods for predictive process control (PPC) of manufacturing processes. More specifically, the subject technology provides improvements to manufacturing processes and specifically includes systems and methods for adaptively controlling various stations in a manufacturing process and optimizing the final manufactured product and the manufacturing process based on predictions made using machine learning models. As discussed in further detail below, some aspects of the technology include systems and methods for training machine learning models. Background Art
[0004] To manufacture products that consistently meet required design specifications safely, on time, and with minimal waste, the manufacturing process needs to be continuously monitored and adjusted. Summary of the invention
[0005] In some aspects, the disclosed technology relates to using a deep learning controller to monitor and improve a manufacturing process. In some embodiments, the disclosed technology includes a computer-implemented method comprising the steps of: receiving a plurality of control values from two or more stations by a deep learning controller, wherein the control values are generated at two or more stations deployed in a manufacturing process; predicting an intermediate output or a final output of a manufactured product based on the control values by the deep learning controller; determining by the deep learning controller whether the predicted intermediate output or final output specification of the manufactured product meets the specification; and if the predicted intermediate output or final output of the manufactured product does not meet the specification, generating one or more control inputs by the deep learning controller, wherein the one or more control inputs are configured to make the intermediate output or the final output of the manufactured product meet the specification.
[0006] In another embodiment, the disclosed technology includes a system having one or more processors and a non-volatile memory storing instructions, which, when executed by the one or more processors, cause the one or more processors to perform operations including: a deep learning controller receiving multiple control values from two or more stations, wherein the control values are generated at the two or more stations deployed in a manufacturing process; the deep learning controller predicting an intermediate output or a final output of a manufactured product based on the control values; the deep learning controller determining whether the predicted intermediate output or final output of the manufactured product meets specifications; and if the predicted intermediate output or final output of the manufactured product does not meet specifications, generating one or more control inputs by the deep learning controller, wherein the one or more control inputs are configured to make the intermediate output or the final output of the manufactured product meet specifications.
[0007] In yet another embodiment, the disclosed technology includes a non-transitory computer-readable storage medium including instructions stored therein, which, when executed by one or more processors, cause the one or more processors to perform operations including the following: a deep learning controller receives multiple control values from two or more stations, wherein the control values are generated at two or more stations deployed in a manufacturing process; the deep learning controller predicts an intermediate output or a final output of a manufactured product based on the control values; the deep learning controller determines whether the predicted intermediate output or final output specifications of the manufactured product meet the specifications; and if the predicted intermediate output or final output of the manufactured product does not meet the specifications, the deep learning controller generates one or more control inputs, wherein the one or more control inputs are configured to make the intermediate output or the final output of the manufactured product meet the specifications. BRIEF DESCRIPTION OF THE DRAWINGS
[0008] In order to describe the manner in which the above and other advantages and features of the present disclosure can be obtained, the principles briefly described above will be described in more detail by reference to specific embodiments illustrated in the accompanying drawings. It is understood that these drawings depict only exemplary embodiments of the present disclosure and should not be considered to limit the scope of the invention as such, and the principles of the present invention are described and explained with additional specificity and detail through the use of the accompanying drawings, in which:
[0009] Figure 1 An example process for controlling a station using statistical process control (SPC) is illustrated.
[0010] Figure 2 An example embodiment of a deep learning controller configured for predictive process control (PPC) is shown.
[0011] Figure 3 An example method for training a deep learning controller is shown.
[0012] Figure 4 An example method for implementing predictive process control is shown.
[0013] Figure 5 An example method for creating a robust data training set using predictive process control is shown.
[0014] Figure 6 An example method for optimizing a manufacturing process using predictive process control is shown.
[0015] Figure 7 A practical application of using predictive process control to optimize a manufacturing process is shown.
[0016] Figure 8 An example method for recording and creating data output is shown.
[0017] Fig. 9 An overall configuration of an embodiment of a deep learning controller that can implement predictive process control is shown. DETAILED DESCRIPTION
[0018] The detailed description set forth below is intended as a description of various configurations of the subject technology and is not intended to represent a unique configuration that can practice the subject technology. The accompanying drawings are incorporated herein and constitute a part of the detailed description. The detailed description includes specific details for providing a more thorough understanding of the subject technology. However, it is clear and obvious that the subject technology is not limited to the specific details set forth herein and can be practiced without these details. In some cases, structures and components are shown in block diagram form to avoid confusing the concept of the subject technology.
[0019] Additional details regarding the use of image classification in manufacturing control are provided by U.S. Provisional Application No. 62 / 836,202, entitled “DEPLOYMENT OF AN IMAGE CLASSIFICATION MODEL ON A SINGLE-BOARD COMPUTER FOR MANUFACTURING CONTROL,” which is incorporated herein by reference in its entirety.
[0020] Additional details related to users of computational models for optimizing assembly / manufacturing operations are provided by U.S. Provisional Patent Application No. 62 / 836,192, entitled “A COMPUTATION MODEL FOR DECISION-MAKING AND ASSEMBLY OPTIMIZATION IN MANUFACTURING,” and U.S. Patent Application No. 16 / 289,422, entitled “DYNAMIC TRAINING FOR ASSEMBLY LINES,” which are incorporated herein by reference in their entireties.
[0021] The manufacturing process is complex and involves processing raw materials by different processing stations (or "stations") until a final product (referred to herein as the "final output") is produced. Each processing station, except the final processing station, receives input for processing and outputs an intermediate output that is passed to a subsequent (downstream) processing station for additional processing. The final processing station receives input for processing and outputs a final output.
[0022] Each processing station may include one or more tools / equipment that perform a set of processing steps on the raw material received (which may be applied to the first processing station or any subsequent station in the manufacturing process) and / or the output received from the previous processing station (which may be applied to any subsequent processing station in the manufacturing process). Examples of processing stations may include, but are not limited to, conveyors, injection molding machines, cutting machines, mold punches, extruders, CNC milling machines, grinders, assembly stations, 3D printers, quality control and verification stations. Example processing steps may include: transporting output from one location to another (performed by conveyors); feeding material into an extruder, melting the material and injecting the material through a mold cavity, where it cools and hardens into the shape of the mold cavity (performed by injection molding presses); cutting material into specific shapes or lengths (performed by cutting machines); pressing material into specific shapes (performed by molding presses).
[0023] In some manufacturing processes, several processing stations can operate in parallel. In other words, a single processing station can send its intermediate output to one or more processing stations (e.g., 1 to N processing stations), and a single processing station can receive intermediate outputs from one to n stations and combine them. In addition, during a single iteration of a manufacturing process, a single processing station can perform the same processing steps or different processing steps on the raw materials or intermediate outputs received, either sequentially or out of sequence.
[0024] The operation of each processing station may be managed by one or more process controllers. In some embodiments, each processing station has one or more process controllers (referred to herein as "station controllers") that are programmed to control the operation of the processing station (the programming algorithm is referred to herein as a "control algorithm"). However, in some aspects, a single process controller can be configured to control the operation of two or more processing stations.
[0025] An operator or control algorithm may provide a station controller with station controller set points (or "set points" or "controller set points" or CSPs) that represent the desired value and / or range of values for each control value. Attributes / parameters that may be measured during station operation and that are related to the tools / equipment / processing steps of a station are control values or station values. If the measured attribute / parameter is also used to control the station, then it is a "control value". Otherwise, the measured attribute / parameter is a "station value". Examples of control values or station values include, but are not limited to, speed, temperature, pressure, vacuum, rotation, current, voltage, power, viscosity, materials / resources used by the station, throughput, downtime, toxic fumes, the type of steps performed at the station, and the order of steps. Whether a measured attribute / parameter is considered a control value or a station value, and whether the measured attribute / parameter is used to control the station or is merely a byproduct of the station's operation, will depend on the particular station, although the examples are the same.
[0026] The control algorithm may also include instructions for monitoring the control value, instructions for comparing the control value to a corresponding set point, and instructions for determining what action to take when the control value is not equal to the corresponding station controller set point (or is not within a defined range of station controller set points). For example, if the measured current temperature value of the station is below the set point, the station controller may send a signal to increase the temperature of the station's heat source until the current temperature value of the station is equal to the set point. Conventional process controllers used to control stations in a manufacturing process are limited because they follow static algorithms (e.g., on / off control, PI control, PID control, lead / lag control) to adjust the set point and specify the actions to be taken when the control value deviates from the set point. Conventional process controllers also have limited, if any, capabilities for analyzing non-controlled values (e.g., environmental conditions (e.g., external temperature, humidity, light, wear of the station), station values, intermediate or final output values, feedback from other processing stations) and dynamically adjusting the set point or control algorithm of the station controller to control the operation of the associated station.
[0027] As used herein, a process value refers to a station value or control value that is aggregated or averaged for an entire series of stations (or a subset of stations) that are part of a manufacturing process. Process values may include, for example, total throughput time, total resources used, average temperature, average speed.
[0028] In addition to station values and process values, various characteristics of the product output (i.e., intermediate output or final output) of the processing station can also be measured, such as temperature, weight, product size, mechanical characteristics, chemical characteristics, optical characteristics and / or electrical characteristics, the number of design defects, the presence or absence of defect types. The various characteristics that can be measured are generally referred to as "intermediate output values" or "final output values". The intermediate output value / final output value can reflect a single measured characteristic of the intermediate output / final output or a total score based on a set of specified characteristics associated with the intermediate / final output, measured and weighted according to a predefined formula.
[0029] Mechanical properties may include hardness, compression, viscosity, density, and weight. Optical properties may include absorption, reflection, transmission, and refraction. Electrical properties may include resistivity and conductivity. Chemical properties may include enthalpy of formation, toxicity, chemical stability in a given environment, flammability (ability to burn), preferred oxidation state, pH (acidity / alkalinity), chemical composition, boiling point, vapor point. The disclosed mechanical, optical, chemical, and electrical properties are exemplary only and are not intended to be limiting.
[0030] like Figure 1 As shown, the values of the intermediate outputs of the processing stations and the values of the final outputs produced by the manufacturing process can be evaluated according to statistical process control (SPC). Using statistical data, SPC tracks, station by station: (1) the values of the intermediate outputs or final outputs produced by the station over a period of time; and (2) the average value of the intermediate output or final output values and the standard deviation from the average value. For example, Figure 1 The intermediate output values for a particular station are shown. Each point represents an intermediate output and its value produced by that station, and the distance of the point from the mean (represented by the black line in the middle) shows how much the intermediate output values for that particular station deviate from the mean. Upper and lower control limits (UCL and LCL) can be defined for a particular station (e.g., one or more standard deviations above or below the mean). Figure 1 The upper and lower control limits are shown to be set at three standard deviations above and below the mean (represented by the dashed lines). The upper and lower control limits are typically narrower than the upper and lower specification limits (USL and LSL) defined for the intermediate / final output values.
[0031] In some aspects, statistics of intermediate outputs and / or final outputs may be used to determine when an artifact is "in specification," i.e., when the output has met certain pre-specified design requirements. Here, meeting specifications may refer to an artifact or a feature of an artifact meeting or exceeding a specified design requirement or set of requirements. For example, an artifact that is considered to be in specification may be an artifact that meets specified statistical requirements, such as acceptable deviations from an ideal value or average value (mean).
[0032] As long as the values of the intermediate output / final output are within the upper and lower control limits, the processing station / the entire process is considered to be in a state of control and no intervention or corrective action is usually taken. Intervention or corrective action is usually taken when the value of the intermediate output / final output exceeds the upper or lower control limit defined for that measurement. However, SPC control has limited impact on improving or optimizing the manufacturing process because intervention / correction only occurs when the upper / lower control limit is exceeded. Adjustments are usually not made when the process is in a state of control. In addition, SPC evaluates a single station individually without considering trends across multiple stations or the impact of multiple stations together on the final product.
[0033] Therefore, new mechanisms are needed to consider the inputs and outputs of each station individually and in conjunction with other inputs and outputs of other stations in the manufacturing process to intelligently and dynamically adjust the inputs to the station controllers to better control the operation of the associated stations. In particular, new mechanisms are needed to predict the inputs that will optimize the manufacturing process to produce a final output that meets specifications. New mechanisms are also needed to predict the inputs to the design and manufacturing process that will improve the final output. In addition, new mechanisms are needed to reduce the variability of the manufacturing process and the final output.
[0034] Aspects of the disclosed technology address the aforementioned limitations of conventional manufacturing processes by providing mechanisms (which may include systems, methods, devices, apparatus, etc.) for incrementally improving the manufacturing process and the resulting manufactured products without interrupting the ongoing manufacturing process (referred to herein as predictive process control). Thus, the disclosed technology can be used to transform and integrate with existing manufacturing systems and infrastructure without causing disruption to the ongoing manufacturing process. Improvements are achieved by providing dynamic control of one or more processing stations in the manufacturing process through conventional station controllers to: (1) consistently produce final output that meets specifications; (2) optimize the design and manufacturing process of the final output; and (3) reduce variability in the manufacturing process and the final output.
[0035] A deep learning controller based on a machine learning / artificial intelligence (AI) model can be used to evaluate control values / station values / process values as well as intermediate output values and final output values, and determine adjustments to station controller inputs. As will be appreciated by those skilled in the art, the machine learning-based techniques may vary depending on the desired implementation without departing from the disclosed techniques. For example, machine learning techniques may use one or more of the following alone or in combination: hidden Markov models, recurrent neural networks, convolutional neural networks (CNNs), deep learning, Bayesian symbolic methods, reinforcement learning, general adversarial networks (GANs), support vector machines, image registration methods, and applicable rule-based systems.
[0036] The machine learning model can also be based on a clustering algorithm (e.g., a mini-batch K-means clustering algorithm), a recommendation algorithm (e.g., a Miniwise hashing algorithm or a Euclidean locality sensitive hashing (LSH) algorithm), and / or anomaly detection algorithms, such as a local anomaly factor. The machine learning model can be based on supervised and / or unsupervised methods.
[0037] As discussed in this article, machine learning models can also be used to determine the processing stations, control values / station values / process values, and intermediate output values (“key influencers”) that have the greatest impact on the final output value, and optimize the manufacturing process by targeting the key influencers.
[0038] Figure 2 An example deep learning controller 218 is shown, which can be configured to control any number (referred to herein as "n") of processing stations in a manufacturing process using predictive process control, as described later in conjunction with Figure 4-7 discussed. Figure 2 , the n processing stations of the manufacturing process are represented by processing stations 222 and 242. The processing stations can be operated in series or in parallel. The general input 236, the empirical prior 239, the functional prior 238 and the input from each of the n stations (e.g., 222 and 242) can be provided to the deep learning controller 218.
[0039] As used herein, functional priors refers to information related to the functions and known limitations of each processing station (individually and collectively) in the manufacturing process. The specifications of the tools / equipment used by the processing stations are considered functional priors. Example functional priors may include, but are not limited to: a screw-driven extruder having a minimum speed and a maximum speed at which the screw can rotate; a temperature control system having a maximum temperature and a minimum temperature that can be achieved based on its heating and cooling capabilities; a pressure vessel having a maximum pressure that can be tolerated before explosion; a flammable liquid having a maximum temperature that can be achieved before combustion. Functional priors may also include the order in which the various stations that are part of the manufacturing process perform their functions. In some embodiments, several stations may run in parallel and their intermediate outputs may be combined at subsequent stations.
[0040] As used herein, empirical priors refer to information obtained through prior experience, for example, in the case of performing the same or similar manufacturing process; running the same or similar stations; producing the same or similar intermediate outputs / final outputs. In some embodiments, the empirical priors may include acceptable final output values or unacceptable final output values. An acceptable final output value refers to a range, upper limit, or lower limit of a final output value in which the final output is considered "in compliance with specifications". In other words, an acceptable final output value describes the parameters of the final output value that meets the design specifications (i.e., meets the specifications). In contrast, an unacceptable final output value refers to an upper / lower limit or range of final output values in which the final output "does not meet specifications" (i.e., describes the parameters of the final output value that does not meet the design specifications). For example, based on prior experience, it may be known that an O-ring used to seal a pipe will only seal if it has certain compression characteristics. This information can be used to determine acceptable / unacceptable compression values for the final output of the O-ring. In other words, all O-ring final outputs with acceptable compression values are able to perform their sealing function, while all O-ring final outputs with unacceptable compression values cannot perform their sealing function. An acceptable intermediate output value (which may be defined by station) refers to an upper / lower limit or range of intermediate output values that defines a parameter of an intermediate output that may ultimately produce a final output that meets specifications without corrective action by other stations. An unacceptable intermediate output value (which may also be defined by station) refers to an upper / lower limit or range of intermediate output values that defines a parameter of an intermediate output that will ultimately result in a final output that does not meet specifications unless corrective action is taken by another station. Similarly, acceptable / unacceptable parameters may be defined for other variables associated with the manufacturing process:
[0041]
[0042]
[0043]
[0044] The empirical priors may also include acceptable and unacceptable manufacturing performance indicators. Manufacturing performance indicators calculate one or more aspects of a manufacturing process over multiple iterations (e.g., production volume for a specified time period, production downtime for a specified time period, number of resources used for a specified time period or a specified final output, percentage of product that does not meet specifications for a specified time period, production volume for a specific operator, material costs associated with a specified number of final outputs).
[0045] Generic inputs as used herein refer to values that are not specific to a particular processing station, but rather to an aspect of the overall manufacturing process, such as date, time of day, ambient temperature, humidity, or other environmental conditions that may affect the manufacturing process, operators, operator skill levels, raw materials used in the process, raw material specifications (e.g., color, viscosity, particle size, and other characteristics specific to the raw materials, etc.), specific batch numbers and costs of raw materials, age of equipment / tools at each station, identification information (e.g., production order number, batch number, lot number, finished product number, and finished product serial number).
[0046] Note that the examples provided for each of functional priors, empirical priors, and generic inputs represent one way to categorize these examples, and other suitable categorizations may be used. For example, another way to categorize the inputs provided to the deep learning controller 218 is: pre-processing inputs (e.g., empirical priors, functional priors, material properties, scheduling requirements); in-process inputs (e.g., generic inputs, control values, station values, intermediate values, final output values, process values); post-processing inputs (e.g., manufacturing performance indicators and other analyses).
[0047] Each processing station may be controlled by one or more associated station controllers (e.g., station controller 220 controls processing station 222, and station controller 240 controls processing station 242). In other embodiments, a single station controller may control multiple processing stations. The deep learning controller 218 may provide control inputs (represented by 226 and 246) based on predictive process control to each processing station controller. In response to received control inputs (e.g., 226 and 246), each station controller may provide one or more control signals (e.g., 221 and 241) that provide commands for adjusting control values (e.g., control values 225 and 245) of the station. Each station outputs an intermediate output (e.g., 224 and 244) having intermediate output values (234a and 244a, respectively). All intermediate output values and final output values from the processing stations are provided to the deep learning controller 218. Each station also outputs station values (e.g., 228 and 248) to the deep learning controller 218. Figure 2 Also shown is that the intermediate output 224 is sent (step 250) to 1 to n subsequent stations, which may represent a single station or a plurality of n stations. Figure 2 As shown) the intermediate input may be received (step 260) from 1 to n preceding stations.
[0048] It should be understood that the communication between the deep learning controller 218, the station controller and the processing station can use any suitable communication technology that provides the ability to communicate with one or more other devices and / or to conduct data transactions with a computer network. For example, the communication technology implemented can include, but is not limited to: analog technology (e.g., relay logic), digital technology (e.g., RS232, Ethernet or wireless), network technology (e.g., local area network (LAN), wide area network (WAN), Internet), Bluetooth technology, near field communication technology, secure RF technology and / or any other suitable communication technology.
[0049] In some embodiments, operator input can be transmitted to the deep learning controller 218 and / or any station controller or processing station using any suitable input device (e.g., keyboard, mouse, joystick, touch, touch screen, etc.).
[0050] In some embodiments, one or more processing stations can be manually operated, such as by a human operator executing specific instructions. The operator (not an electronic station controller) follows a set of instructions, which can be provided manually or electronically (e.g., via a video or computer display). For example, at a manual station, an operator can perform the functions of cutting a wire to a specific length and measuring the length of the cut wire. Manual feedback, such as the length of the cut wire, can be provided to the deep learning controller 218. As described herein, using predictive process control, the deep learning controller 218 can determine whether the wire is cut to the desired length specification and provide improvements to, for example, the cutting process, which can be provided to the operator of the manual station in the form of a set of instructions.
[0051] Figure 3 A process 300 for tuning (training) a deep learning controller 218 is provided in accordance with some embodiments of the disclosed subject matter.
[0052] In step 310, conventional methods may be used to initialize the set points, algorithms, and other control inputs for each station controller in the manufacturing process. In addition, the control algorithm / operator may provide initial control values / station values. The control algorithm, initial set point values, and initial control values / station values may be provided to the deep learning controller 218 (step 315).
[0053] Note that control values, control algorithms, set points, and any other information provided to the station controller (e.g., process timing, equipment commands, alarm alerts, emergency stops) will be collectively referred to as "station controller inputs" or "control inputs."
[0054] Additionally, other inputs may be provided to the deep learning controller 218 , such as functional priors 238 , empirical priors 239 , and generic inputs 236 .
[0055] In step 325, the manufacturing process is iterated to all processing stations using conventional control methods. As described above, the processing stations discussed herein may be operated in series or in parallel. In addition, a single station may perform: a single processing step multiple times (sequentially or non-sequentially), or different processing steps (sequentially or non-sequentially) for a single iteration of the manufacturing process. The processing station generates intermediate outputs, and if it is a final station, a final output. The intermediate outputs are transmitted to subsequent (downstream) stations in the manufacturing process until the final output is produced.
[0056] As the manufacturing process iterates to each processing station, all values associated with: a single station (e.g., control value), the output of a single station (e.g., station value, intermediate output value / final output value), or the output of multiple stations (e.g., process value) are measured or calculated; and the values are provided to adjust the machine learning algorithm of the deep learning controller 318 (steps 327 and 328).
[0057] In some embodiments, manufacturing performance indicators for a manufacturing process under conventional control (e.g., production volume for a specified time period, production downtime for a specified time period, number of resources used for a specified time period or a specified final output, percentage of products that do not meet specifications for a specified time period, production volume of a specific operator, material costs associated with a specified number of final outputs) can be calculated and provided to the deep learning controller 218 (step 329).
[0058] Although not shown, any actions taken (or control signals generated) by the station controller in response to received control values or other control inputs from the processing station may be provided to the deep learning controller 218. Such actions may include adjusting temperature, speed, etc. In addition, deviations from acceptable set points, acceptable intermediate output values / final output values, and acceptable control values / station values / process values may also be calculated and provided to the deep learning controller 218.
[0059] Note that all inputs to the deep learning controller 218 may be entered electronically or manually by an operator.
[0060] The adjustment of the machine learning model of the deep learning controller 218 (step 235) can be implemented by an unsupervised learning method. In addition to the functional priors 238, empirical priors 239, and general inputs 236 input to the deep learning controller 218, the deep learning controller 218 simply makes inferences by analyzing the received data collected during the iteration of the manufacturing process (e.g., steps 328 and 329). In other embodiments, the deep learning controller 218 can be adjusted by a supervised learning method, a combination of supervised and unsupervised methods, or a similar machine learning method. In addition, the training of the deep learning controller 218 can be enhanced by providing the deep learning controller 218 with simulated data or data from a similar manufacturing process. In one embodiment, the deep learning controller 218 can be adjusted by implementing the deep learning controller 218 into a similar manufacturing process and fine-tuning the deep learning controller during the implementation of the target manufacturing process. That is, the training of the deep learning controller 218 can be performed using a training process performed before the deep learning controller 218 is deployed in the target manufacturing environment.
[0061] Based on adjustments to the machine learning model, the deep learning controller 218 can predict values of features of the final output (“expected values” or “EVs”), which determine whether the final output values are acceptable (i.e., whether the final output is “meets specifications”) (step 342). The deep learning controller 218 can provide a confidence level in its predictions at a certain moment in time or over a specific time period, for example, to provide a measure of statistical confidence in the predictions. In some aspects, the confidence level can be expressed as a numerical probability of the accuracy of the prediction, and in other aspects, the confidence level can be expressed as an interval or probability range. In step 343, the deep learning controller 218 can compare the expected value with the actual measured value of the specified feature of the final output (“actual value” or “AV”).
[0062] In some aspects, the deep learning controller 218 can be configured to perform EV predictions on output features on a station-by-station basis. That is, the deep learning controller 218 can make EV predictions for the outputs of a particular station and then compare those predictions with the actual outputs observed at that station. Alternatively, EV predictions can be made for the outputs resulting from combined processing performed by two or more stations, depending on the desired implementation.
[0063] As the manufacturing process progresses at each station and the deep learning controller 218 receives additional information, the deep learning controller 218 can modify its expectations and confidence levels. If the predictions of the deep learning controller 218 are correct at a predefined threshold confidence level within a specified time period, the deep learning controller 218 can provide a signal that the deep learning controller 218 is ready to control the operation of the processing station.
[0064] In some embodiments, the deep learning controller 218 may also predict whether any control input will result in unsatisfactory station performance or affect process performance (i.e., result in unacceptable process performance) after initialization of the station controller, at the beginning of an iteration of the manufacturing process (i.e., after the manufacturing process is executed at all stations), and throughout the manufacturing process. The deep learning controller 218 may provide a confidence level in its prediction. The deep learning controller 218 may determine whether the prediction of the deep learning controller 218 is correct. In further embodiments, if the prediction of the deep learning controller 218 (regarding the expected final output value and the predicted station / process performance) is correct within a specified time period and with a threshold confidence level defined by the operator, the deep learning controller 218 may provide a signal that the deep learning controller 218 is ready to control the operation of the processing station.
[0065] Figure 4 An example process for controlling a manufacturing process using predictive process control is shown.
[0066] The deep learning controller 218 uses its adjusted machine learning algorithm (such as combined Figure 3 The deep learning controller 218 may calculate the control inputs for the station controllers associated with the processing stations of the manufacturing process (as discussed above). Based on the calculated control inputs, the deep learning controller 218 may predict the expected value (EV) of the final output of the manufacturing process and the confidence level for the prediction (step 405). If the deep learning controller 218 determines that the expected value will meet the specifications at a threshold confidence level (step 415), the deep learning controller 218 may output the calculated control inputs to the station controllers associated with the processing stations of the manufacturing process (step 420). The deep learning controller 218 may calculate the control inputs and provide the calculated control inputs to one or more station controllers at the beginning of the manufacturing process or continuously throughout the manufacturing process. The control inputs do not have to be provided to the station controllers in a serial order, but may be provided to one or more station controllers in parallel or in any order suitable for producing a final output that meets the specifications. Each station controller that receives a control input from the deep learning controller 218 may send a control signal to its associated station to control the control value (e.g., control the control value so that they match the received set point). The machine learning algorithm is continuously improved throughout the implementation of the PPC (step 335). Furthermore, functional priors and empirical priors can be dynamically updated throughout PPC.
[0067] At step 430, the manufacturing process is performed at all processing stations, either serially or in parallel. As the process traverses each station, all values associated with: a single station (e.g., control values); the output of a single station (e.g., station value, intermediate output value / final output value) or the output of multiple stations (e.g., process values) can be measured or calculated and provided to adjust the machine learning algorithm of the deep learning controller 218 (step 432). In addition, manufacturing performance indicators of the manufacturing process under predictive process control can be calculated and provided to the deep learning controller 218 (step 432). The process values and manufacturing performance indicators calculated under PPC can be compared with the process values and manufacturing performance indicators calculated under conventional control to determine the improvements provided by predictive process control.
[0068] The whole process is like Figure 4 As shown, the deep learning controller 218 can predict an expected value (EV) of the final output, determine whether the expected value of the final output meets specifications, determine a confidence level for the prediction, and then provide feedback about its prediction by comparing the expected final value to the actual final value (step 445). In addition, if the deep learning controller 218 determines that the final output does not meet specifications, adjustments to the control inputs can be calculated so that the expected value of the predicted final output meets specifications.
[0069] In some aspects, the deep learning controller 218 can be configured to perform EV predictions on the intermediate outputs on a station-by-station basis. That is, the deep learning controller 218 can make EV predictions for the outputs of a particular station, determine whether the EV of the intermediate outputs meets specifications, determine a confidence level for its predictions, and then compare these predictions with the actual outputs observed at that station. Alternatively, EV predictions can be made for the outputs resulting from combined processing performed by two or more stations, depending on the desired implementation. In addition, if the deep learning controller 218 determines that the intermediate outputs do not meet specifications, adjustments to the control inputs can be calculated so that the expected value of the predicted intermediate outputs meets specifications.
[0070] Note that if the confidence level determined by the deep learning controller 218 is below a predetermined threshold, control of the manufacturing process can revert to conventional control, such as in conjunction with Figure 3 as described.
[0071] In some embodiments, the deep learning controller 218 may also monitor whether any of the station value / control value / process value or intermediate output value is unacceptable and make further adjustments to the station controller input, or generate an alarm if the problem cannot be resolved by adjusting the station controller input.
[0072] Based on the data received while the station is running throughout the manufacturing process, the deep learning controller 218 can adjust the control inputs of one or more station controllers. In a further embodiment, the deep learning controller 218 can not only initialize the station controller inputs before the iteration of the processing station of the manufacturing process begins, but also adjust the station controller inputs during the manufacturing process itself ("feedforward control"). Specifically, based on the information received from the previous station in the manufacturing process, the deep learning controller 218 can make changes to the control inputs associated with the subsequent stations in the process. For example, if the deep learning controller 218 determines that the intermediate output of a particular station is defective, the deep learning controller 218 can determine whether there are any corrective actions that can be taken at the subsequent station so that the final output meets the specifications. The deep learning controller 218 can also change the current and previous processing stations in real time based on feedback about control values / station values / process values and / or intermediate output values / final output values ("feedback control"). This ability to dynamically control each station in real time and adjust the downstream station controller to compensate for errors, miscalculations, undesirable conditions, or unforeseen results occurring upstream increases the likelihood of manufacturing a final output that meets specifications. Furthermore, even though a wide range of final output values may be considered within specification, it may be desirable to manufacture final outputs of the same or similar quality and with similar final output values that fall within a narrower range of acceptable final output values. Feedback and feedforward control and the predictive capabilities of the deep learning controller enable the deep learning controller 218 to adjust the station controllers to produce final output values of consistent quality and similar final output values.
[0073] It is useful to identify which parameters of a manufacturing process have the greatest impact on the final output value or process performance ("key influencers"). The deep learning controller 218 can consider all parameters of the manufacturing process (e.g., one or more control values, one or more station values, one or more process values, one or more stations, one or more intermediate outputs, or any combination thereof), and using one or more of its machine learning algorithms can identify key influencers. In some aspects, the deep learning controller 218 can employ unsupervised machine learning techniques to discover one or more key influencers, for example, where each key influencer is associated with one or more parameters (or combinations of parameters) that affect the characteristics of various station outputs, final outputs, and / or process performance. It should be understood that key influencers and their associated parameters can be discovered through operation and training of the deep learning controller 218 without explicitly marking, identifying, or otherwise outputting key influencers or parameters to an operator.
[0074] In some approaches, the deep learning controller 218 can rank the impact of each parameter of the manufacturing process on the final output value or process performance in order of importance. Key influencing factors can be identified based on the following factors: cutoff ranking (e.g., the top 5 aspects of the manufacturing process that affect the final output value), the lowest degree of impact (e.g., all aspects of the manufacturing process that contribute at least 25% to the final output value); or any other suitable criteria. In some aspects, key influencing features can be associated with quantitative scores, for example, related to the impact weight of the corresponding feature.
[0075] The deep learning controller 218 may continuously calculate key influencing factors throughout the manufacturing process (step 446 ).
[0076] In some embodiments, such as in combination Figure 5 As described, key influencing factors may be used to help build a more robust dataset for training the deep learning controller 218 .
[0077] In conventional manufacturing, the data generated from the manufacturing process is limited because the goal is to produce intermediate output values within a specified standard deviation of the mean. Therefore, the range of control inputs, control values / station values / process values is also limited because they are all designed to produce intermediate output values that fall within a specified range of the mean. In contrast, according to the present disclosure, as long as the final output meets the specification, the intermediate output value does not necessarily fall within the specified range of the mean. As the predictions of the deep learning controller become more accurate, in some embodiments, the deep learning controller 218 can intentionally change the control input to form conditions for producing intermediate output values that may exceed the normal fluctuations of the control process under traditional manufacturing processes (e.g., SPC), but still generate a final output that meets the specification. This creates a more robust data training set for the deep learning controller 218 to detect patterns and determine how specific stations, station values / control values / process values, and intermediate output values affect the final output value (e.g., whether the final output meets the specification). Robust data sets can be created both during and before the implementation of the deep learning controller 218 in a production environment.
[0078] Figure 5An example process for creating a more robust data set is shown. In some embodiments, the deep learning controller 218 can adjust the known controller inputs (e.g., control set points) of one or more station controllers to produce intermediate output values that may exceed a specified range of the average value. For example, once the deep learning controller 218 is adjusted (step 335), it knows at least some of the control inputs of each station controller that will result in a final output value that meets the specifications. In some embodiments, the deep learning controller 218 can select one or more station controllers and cause the known control inputs (e.g., set points) of the selected station controllers to change by a predetermined threshold value (e.g., new set point = original set point + 1% original set point) (step 510). The deep controller 218 can use the newly calculated control inputs and the confidence level for its prediction to predict the expected value (EV) of the final output of the manufacturing process (step 515). If the deep learning controller 218 determines with a threshold confidence level that the expected value will meet the specifications (step 517), the deep learning controller 218 can provide the adjusted control input to the selected station controller (step 520). The deep learning controller 218 may also compare the predicted desired value to the actual final output value (AV) to provide feedback on its prediction and further adjust the control input (step 525).
[0079] In one example embodiment, control inputs related to material tolerances may be purposefully changed, and the deep learning controller 218 may be used to determine which adjustments to make to other control inputs to produce a final product that meets specifications. By training the deep learning controller 218 in this manner, when new materials are unexpectedly introduced into the manufacturing process, the deep learning controller 218 may adjust the control inputs on its own, without operator input. Similarly, adjustments may be purposefully made to the control algorithm (e.g., adjustments that mimic a virus or other attack), and the deep learning controller 218 may be used to determine which adjustments to make to other control inputs to produce a final product that meets specifications. By purposefully introducing these changes, when adjustments are unexpectedly made to the control algorithm during the manufacturing process, the deep learning controller 218 may adjust the control inputs on its own, without operator input.
[0080] In some embodiments, the deep learning controller 218 may first determine the key influencing factors (such as Figure 4Step 446 described above) and changes the control input associated with the key influencing factor. In other embodiments, the deep learning controller 218 may change the control inputs of all station controllers, or select station controllers according to a predetermined formula. In order to create a robust data set, the deep learning controller 218 may continue to adjust the control input associated with a particular station controller at each iteration of the manufacturing process (step 528). For example, the deep learning controller 218 may adjust the set point associated with the key influencing factor by 1% and iterate one or more times during the manufacturing process. In subsequent iterations, the deep learning controller 218 may adjust the set point associated with the key influencing factor by another 1% and iterate one or more times during the manufacturing process. The deep learning controller 218 may continue to make adjustments until the expected value of the manufacturing process when using the adjusted set point will result in a final output value that meets the specification at a threshold confidence level. During each iteration, as combined with Figure 4 As described, station values / control values, intermediate output values / final output values, process values, and manufacturing performance indicators are generated (step 430) and used to adjust the machine learning algorithm of the deep learning controller 218 (step 335) and dynamically update functional priors and empirical priors.
[0081] The following example further illustrates the creation of a robust data set by varying a control input (e.g., a temperature set point for a particular station). In this example, the set point temperature for a particular station is 95°, and the actual station temperature fluctuates between 92° and 98° (i.e., 3° above and below the set point temperature). Both the set point temperature of 95° and the corresponding ±3° fluctuations in the actual station temperature produce final output values that meet specifications. The deep learning controller 218 can then predict whether adjusting the temperature set point by a negligible amount (e.g., ±.5°) will still produce a final output value that meets specifications. If the deep learning controller 218 predicts that the final output value of the manufacturing process using the adjusted temperature set point meets specifications at a threshold confidence level, the deep learning controller 218 adjusts the temperature set point by ±.5°. Assuming the same station temperature fluctuation of ±3° as the set point, the actual station temperature will be in the range between 92.5° and 98.5° when the set point is 95.5°; and between 91.5° and 97.5° when the set point is 94.5°. The deep learning controller 218 can compare the final output value to the expected output value within the temperature range of 91.5° to 98.5° and determine whether it correctly predicted that the final output value will meet specification. Because the temperature set point is adjusted by ±.5°, the resulting data set covers a wider temperature range than the original temperature range of 92°-98°: 91.5°-98.5.
[0082] Note that instead of changing the station controller set point, the control value of the station can be changed by modifying other control inputs to the station controller (e.g., control values), which will achieve the same goal of changing the set point. For example, assume the following: the set point of the station controller is 100 degrees, the actual station temperature value (i.e., control value) is 100 degrees, and the goal is to increase the station actual temperature value by 2 degrees. Instead of increasing the temperature set point to 102 degrees to achieve this change, the deep learning controller 218 can change the control value it provides to the station controller to be two degrees lower than the actual temperature value (e.g., changing the control value from 100 degrees to 98 degrees), thereby causing the station controller to increase the station temperature by two degrees to increase the ideal station temperature by two degrees (i.e., 102 degrees). When the existing station control does not allow the set point to be changed, it may be necessary to change the control value instead of the set point.
[0083] In some embodiments, such as in combination Figure 6 The described, key influencing factors can be used to optimize the final output value or process value.
[0084] Once the most important stations, station values / control values / process values, and intermediate output values that affect the final output value or process performance are identified, resource allocation and process optimization can be performed for key influencing factors. For example, data resources (e.g., collection, processing, and storage) can be allocated primarily to data related to key influencing factors ("curated data"), rather than collecting large amounts of data that have a minor impact on the manufacturing process. In addition, curated data (a subset of all available data in the manufacturing process) can be provided to machine learning algorithms to optimize key influencing factors, thereby reducing the number of training examples and increasing the resources available to process curated data. In addition, machine learning algorithms can optimize for key influencing factors rather than the entire process, thereby reducing the possible states and actions that the machine learning algorithm must consider, and allocating resources more efficiently and intelligently. For example, in reinforcement learning, the state and action space defines the possible states that the agent may perceive and the range of actions available to the agent. By using reinforcement learning only for key influencing factors, the state and action space is reduced, making the algorithm easier to manage.
[0085] In some embodiments, the operator may specify one or more characteristics of the final output or process values that it wants to optimize (e.g., using the least power or resources, fastest throughput, least number of defects, maximum tensile strength). The deep learning controller 218 may run a machine learning model such as reinforcement learning to optimize the specified characteristics of the final output (“optimal design value”) and / or the specified process value (“optimal process value”) for the key influencing factors.
[0086] Figure 6Some embodiments according to the subject disclosure illustrate processes for specifying optimal design / process values and using a deep learning controller to control and optimize key influencing factors to achieve the desired optimal design or process values.
[0087] At 600, the desired optimal design / process values are provided to the deep learning controller 218. For example, generating a final output that meets specifications: using the least power or resources, fastest throughput, least number of defects, greatest tensile strength, etc.
[0088] As shown at 605, the deep learning controller 218 can determine the key influencing factors for driving a product that meets specifications (see also Figure 4 ), and can predict control inputs ("optimal control inputs") to control each key influencing factor to achieve the desired optimal design or process value. The deep learning controller 218 can use the optimal control inputs to determine the expected value of the final output of the manufacturing process, and predict whether the expected value meets the specifications and achieves the desired optimal design or process value (step 615). The deep learning controller 218 can also calculate the confidence level that the optimal control input will lead to the optimal design or process value (step 615). The deep learning controller 218 can provide optimal control inputs to the relevant station controllers at the beginning and continuously throughout the manufacturing process (step 620). The optimal control inputs do not have to be provided to the station controllers in a serial order, but can also be provided to one or more station controllers in parallel or in any order suitable for producing the final output to achieve the desired optimal design or process value.
[0089] In other embodiments, the parameters of the final output that meets the specifications will be updated to match the desired optimal design / process values. For example, if the tensile strength of the final output that meets the specifications is 40-90 megapascals (MPa), and the optimal tensile strength of the final output is determined to be 70-90 MPa, the parameters that meet the specifications can be updated to 70-90 MPa. The deep learning controller 218 can predict that the calculated control inputs of the station controllers associated with the key influencing factors will achieve the updated specification parameters (i.e., 70-90 MPa) and determine the confidence level. When the deep learning controller 218 predicts that the calculated control inputs will achieve the updated specification parameters with a confidence level above a predefined threshold, the deep learning controller 218 can update the relevant station controllers.
[0090] Note that the deep learning controller 218 can continue to provide control inputs to other station controllers, such as in conjunction with Figure 3 As discussed, the optimization will only be for station controllers associated with key influencing factors. In some embodiments, the deep learning controller 218 can optimize station controllers, station values / control values / process values, or intermediate output values associated with any station, rather than just optimizing for key influencing factors.
[0091] The deep learning controller 218 may also compare the expected values with the actual values to provide feedback on its predictions and further adjust the control inputs (step 545). When the manufacturing process begins and proceeds to all stations, in some embodiments, only the measurements related to the key influencing factors and the measurements of the final output will be collected and provided to the deep learning controller 218. In other embodiments, the deep learning controller 218 may continue to collect data from all processing stations. This data will be used to continuously improve the control inputs of the key influencing factors so that the optimal design / process values are consistently achieved with high confidence.
[0092] Note: The desired optimal design / process values can be changed at any time. In addition, new key influencers can be calculated by changing the criteria that may be used as key influencers (for example, a first criterion may classify the top 5 stations that drive a final output that meets specifications as key influencers, while an updated criterion may classify only the top 3 stations that drive a final output that meets specifications as key influencers). In addition, once the deep learning controller 218 predicts the control input to achieve the optimal design / process values over a period of time with a certain confidence level, the deep learning controller 218 can identify which key influencers are the key influencers driving the desired optimization and only target a subset of the key influencers, further reducing the possible actions / states that the machine learning algorithm must consider and more efficiently allocating resources to that subset.
[0093] In further embodiments, the deep learning controller 218 may not only initialize station controller inputs for key influencing factors before an iteration of a station of a manufacturing process begins, but may also adjust control inputs during the process itself. In particular, based on information received from previous stations in the process, the deep learning controller 218 may change control inputs associated with subsequent stations in the manufacturing process to ensure that optimal design / process values are achieved. The deep learning controller 218 may also adjust previous stations in the manufacturing process and receive data from subsequent stations as the manufacturing process proceeds.
[0094] In some embodiments, manufacturing performance indicators for the manufacturing process attempting to achieve the optimal design or process values may be calculated and provided to the deep learning controller 218 (step 632). These manufacturing performance indicators may be compared with manufacturing performance indicators of the PPC and / or conventionally controlled manufacturing performance indicators to determine the improvements provided by optimizing the design and / or manufacturing process.
[0095] Examples of optimizing a manufacturing system to reduce the possible actions and states that can be adjusted and implemented by a deep learning controller are described in U.S. Provisional Application No. 62 / 836,199, entitled “Adaptive Methods and Real-Time Decision Making for Manufacturing Control,” which is incorporated herein by reference in its entirety. The disclosed methods are examples only and are not limiting.
[0096] Examples of optimizing a manufacturing system to reduce the possible actions and states that a machine learning algorithm must consider and that are regulated and implemented by a deep learning controller are described in U.S. Provisional Application No. 62 / 836,213, entitled “TRANSFER LEARNING APPROACH TO MULTI-COMPONENT MANUFACTURING CONTROL,” which is incorporated herein by reference in its entirety. The disclosed methods are exemplary only and not limiting.
[0097] Figure 7 An example manufacturing system is shown in which predictive process control can be applied to optimize the design / processing of a final output. Specifically, a 3D manufacturing system can have multiple processing stations (e.g., stations 700-750), and each processing station can perform a processing step (e.g., depositing a material extrusion layer that will produce a final output). Using predictive process control, such as Figure 6 As shown, the operator can specify what he wants to optimize (e.g., the tensile strength of the final output) (step 600). In addition, the deep learning controller 218 can determine the key influencing factors that affect the tensile strength of the final output (step 605) (e.g., the volume change of each layer of deposited material and the change in the extruder speed control). The deep learning controller 218 can reduce the possible actions / states that the machine learning algorithm must consider by running a machine learning algorithm (e.g., reinforcement learning) to optimize only the key influencing factors, without optimizing other control properties (e.g., extruder nozzle temperature, printing mode of the printer head). In addition, the deep learning controller 218 can identify which extrusion layers have the greatest impact on tensile strength (e.g., layers 4 and 5), and optimize the key influencing factors for the stations that deposit these layers (e.g., stations 4 and 5).
[0098] Note that in some embodiments, the combination Figure 4-7 With the predictive process control method discussed in , the conventional station controller can be shut down and the deep learning controller 218 can directly control the processing station.
[0099] Additionally, in some embodiments of predictive process control, the data logging module 810 (e.g. Figure 8815). For example, the data logging module 810 may be programmed to search the received data for predefined trigger events and generate reports, emails, alarms, log files, or other data outputs displaying relevant data associated with those trigger events (step 815). For example, adjusting a set point or other control input may be defined as a trigger event, and the following data may be reported: the name of the set point or other control input adjusted, the station affected, the date and time, the confidence level associated with the adjusted control input, whether the adjusted control input achieved the final output that met the specifications or the optimal design / process value, and the reason for the adjustment. In another example, achieving or not achieving the optimal design / process value may be defined as a trigger event, and the following data may be reported: the control input of the key influencing factor, the date and time, any reported outages, the confidence level of the control input, the production time, and the operator related to resource consumption. Other suitable triggers may be defined, and other suitable data may be reported. The data logging module can also compare the process values, final output values, and manufacturing performance indicators collected during the manufacturing process using traditional control with the process values, final output values, and manufacturing performance indicators collected during the manufacturing process using predictive performance control. In some embodiments, the data logging module 810 can be included in the deep learning controller 218.
[0100] Fig. 9 A general configuration of an embodiment of a deep learning controller 218 that can implement predictive process control according to some embodiments of the disclosed subject matter is shown. Although the deep learning controller 218 is shown as a localized computing system in which various components are coupled via a bus 905, it should be understood that the various components and functional computing units (modules) can be implemented as separate physical or virtual systems. For example, one or more components and / or modules can be implemented in a physically separate remote device, such as using a virtual process (e.g., a virtual machine or container) instantiated in a cloud environment.
[0101] The deep learning controller 218 may include a processing unit (e.g., one or more CPUs and / or one or more processors) 910 and a bus 905 that couples various system components including a system memory 915 (e.g., a read-only memory (ROM) 920 and a random access memory (RAM) 925) to the processing unit 910. The processing unit 910 may include one or more processors, such as processors from the Motorola family of microprocessors or the MIPS family of microprocessors. In alternative embodiments, the processing unit 910 may be specially designed hardware for controlling the operation of the deep learning controller 218 and performing predictive process control. When acting under the control of appropriate software or firmware, the processing module 910 may perform the calculations of the various machine learning algorithms and PPC described herein.
[0102] The memory 915 may include various types of memory with different performance characteristics, such as a memory cache 912. The processor 910 may be coupled to a storage device 930, which may be configured to store software and systems required to implement one or more functional modules and / or database systems. Each of these modules and / or database systems may be configured to control the processor 910 as well as a dedicated processor in which the software instructions are incorporated into the actual processor design.
[0103] To enable the operator to interact with the deep learning controller 218, the input device 945 can represent any number of input mechanisms, such as a microphone for voice, a touch-sensitive screen for gesture or graphic input, a keyboard, a mouse, motion input, and the like. The output device 935 can also be one or more of a plurality of output mechanisms known to those skilled in the art (e.g., a printer, a monitor). In some cases, a multi-mode system can enable an operator to provide multiple types of input to communicate with the deep learning controller 218. The communication interface 940 can generally govern and manage operator input and system output, as well as receive all electronic inputs from and send them to other components that are part of the manufacturing process (e.g., station controllers, processing stations, data logging modules, and all related sensors and image capture devices) as part of the manufacturing process. There is no limitation on the operation on any particular hardware arrangement, so the basic functionality here can be easily replaced with improved hardware or firmware arrangements as the hardware or firmware arrangements are developed. The data output from the deep controller 218 can be visually displayed, printed, or generated in the form of a file and stored in the storage device 930 or transmitted to other components for further processing.
[0104] The communication interface 940 may be provided as an interface card (sometimes referred to as a "line card"). Typically, they control the sending and receiving of data packets over the network and sometimes support other peripheral devices used with the router. Interfaces that may be provided include Ethernet interfaces, frame relay interfaces, cable interfaces, DSL interfaces, token ring interfaces, and the like. In addition, various ultra-high-speed interfaces may be provided, such as fast token ring interfaces, wireless interfaces, Ethernet interfaces, Gigabit Ethernet interfaces, ATM interfaces, HSSI interfaces, POS interfaces, FDDI interfaces, and the like. Typically, these interfaces may include ports suitable for communicating with appropriate media. In some cases, they may also include independent processors and, in some cases, volatile RAM. Independent processors may control communication-intensive tasks such as packet switching, media control, and management. By providing separate processors for communication-intensive tasks, these interfaces allow the processing unit 910 to efficiently perform machine learning and other calculations required to achieve predictive process control. The communication interface 940 may be configured to communicate with other components that are part of the manufacturing process, such as station controllers, processing stations, data logging modules, and all associated sensors and image capture devices.
[0105] Sensors associated with the manufacturing process may include sensors that existed before the implementation of PPC, as well as any new sensors that are added to perform any additional measurements used by PPC. One or more sensors may be included in or coupled to each station. Sensors can be used to measure values generated by the manufacturing process, such as: station values, control values, intermediate output values, and final output values. In addition, the deep learning controller 218 or modules external to the deep learning controller 218 can use the information provided by the sensors to calculate process values and manufacturing performance indicators. Example sensors may include, but are not limited to: rotary encoders for detecting position and speed; sensors for detecting proximity, pressure, temperature, liquid level, flow, current, and voltage; limit switches for detecting states (such as whether there is a travel end limit). As used herein, sensors include both sensing devices and signal conditioners. For example, a sensing device reacts to a station value or a control value, and a signal conditioner converts the reaction into a signal that can be used and interpreted by a deep learning controller. Examples of sensors that react to temperature are RTDs, thermocouples, and platinum resistance probes. Strain sensors react to pressure, vacuum, weight, distance changes, and the like. A proximity sensor reacts to an object when the object is a certain distance away from each other or within a specified distance. For all of these examples, the reaction must be converted into a signal that can be used by the deep learning controller 218. In many cases, the signal conditioning function of the sensor produces a digital signal that is interpreted by the deep learning controller 218. The signal conditioner can also produce an analog signal or a TTL signal, etc.
[0106] In some embodiments, the deep learning controller 218 may include an imaging processing device 970 that processes images received by various image capture devices (e.g., cameras) that are coupled to one or more processing stations and are capable of monitoring and capturing images of intermediate and final outputs. These images may be transmitted to the deep learning controller 218 via the communication interface 940 and processed by the image processing device 970. The images may be processed to provide data, such as the number and type of defects, output size, throughput, which may be used by the deep learning controller 218 to calculate intermediate and final output values. In some embodiments, the image processing device may be external to the deep learning controller 218 and provide information to the deep learning controller 218 via the communication interface 940.
[0107] The storage device 930 is a non-volatile memory and may be a hard disk or other type of computer-readable medium that can store computer-accessible data, such as a magnetic tape, a flash memory card, a solid-state storage device, a digital versatile disk, a cassette, a random access memory (RAM) 825, a read-only memory (ROM) 820, and a mixture thereof.
[0108] In practice, the storage device 930 can be configured to receive, store and update input data input to the deep learning controller 218 and output data output from the deep learning controller 218, such as functional priors, empirical priors, general inputs, pre-processed inputs, in-process inputs and post-processed inputs.
[0109] In some embodiments, any suitable computer-readable medium may be used to store instructions for performing the functions and / or processing described herein. For example, in some embodiments, the computer-readable medium may be transient or non-transient. For example, non-transient computer-readable media may include media such as non-transient magnetic media (e.g., hard disks, floppy disks, etc.), non-transient optical media (e.g., optical disks, digital video disks, Blu-ray disks, etc.), non-transient semiconductor media (e.g., flash memory, electrically programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), etc.), any suitable medium that is not fleeting during transmission and does not have any persistence, and / or any suitable tangible medium. As another example, transient computer-readable media may include signals on a network; signals in wires, conductors, optical fibers, circuits, any suitable medium that is fleeting during transmission and does not have any persistence, and / or any suitable intangible medium.
[0110] The various systems, methods and computer-readable media described herein can be implemented as part of a cloud network environment. As used herein, a cloud-based computing system is a system that provides virtualized computing resources, software and / or information to client devices. Computing resources, software and / or information can be virtualized by maintaining centralized services and resources that edge devices can access through communication interfaces such as networks. The cloud can provide various cloud computing services through cloud elements, such as software as a service (SaaS) (e.g., collaboration services, email services, enterprise resource planning services, content services, communication services, etc.), infrastructure as a service (IaaS) (e.g., security services, network services, system management services, etc.), platform as a service (PaaS) (e.g., Web services, streaming media services, application development services, etc.), and other types of services, such as desktop as a service (DaaS), information technology management as a service (ITaaS), hosted software as a service (MSaaS), mobile backend as a service (MBaaS), etc.
[0111] The provision of examples described herein (and the phrases "for example," "such as," "including," etc.) should not be construed as limiting the claimed subject matter to the specific examples; rather, these examples are merely intended to illustrate some of the many possible aspects. One of ordinary skill in the art will understand that the term mechanism may include hardware, software, firmware, or any suitable combination thereof.
[0112] Unless otherwise apparent from the above discussion, it should be understood that throughout the description, discussions using terms such as "determining," "providing," "identifying," "comparing," etc. refer to the actions and processes of a computer system or similar electronic computing device that operates and converts data represented as physical (electronic) quantities in a computer system memory or register or other such information storage, transmission, or display device. Certain aspects of the present disclosure include processing steps and instructions described herein in algorithmic form. It should be noted that the processing steps and instructions of the present disclosure can be embodied in software, firmware, or hardware, and when embodied in software, can be downloaded to and run by different platforms used by real-time network operating systems.
[0113] The present disclosure also relates to a device for performing the operations herein. The device may be specially constructed for the desired purpose, or it may include a general-purpose computer selectively activated or reconfigured by a computer program stored on a computer-readable medium, which the computer can access. Such a computer program may be stored in a computer-readable storage medium, such as, but not limited to, any type of disk, including a floppy disk, an optical disk, a CD-ROM, a magneto-optical disk, a read-only memory (ROM), a random access memory (RAM), an EPROM, an EEPROM, a magnetic card or an optical card, an application-specific integrated circuit (ASIC), or any type of non-transitory computer-readable storage medium suitable for storing electronic instructions. In addition, the computer mentioned in the specification may include a single processor or may be an architecture that uses a multi-processor design to increase computing power.
[0114] The algorithms and operations presented here have no inherent relationship with any specific computer or other device. Various general-purpose systems can also be used together with the programs according to the teachings here, or it can be proved convenient to construct more specialized devices to perform the required method steps and system-related actions. The required structures of various these systems and equivalent variations are obvious to those skilled in the art. In addition, the present disclosure is not described with reference to any specific programming language. It should be understood that a variety of programming languages can be used to implement the teachings of the present disclosure as described herein, and any reference to a specific language is provided for realizing the best implementation of the present disclosure and the present disclosure.
[0115] The logical operations of various embodiments are implemented as: (1) a sequence of computer-implemented steps, operations, or processes running on programmable circuits within a general-purpose computer, (2) a sequence of computer-implemented steps, operations running on special-purpose programmable circuits; and / or (3) machine modules or program engines interconnected within programmable circuits. System 300 can practice all or part of the method, can be part of the system, and / or can operate according to instructions in the non-transitory computer-readable storage medium. Such logical operations can be implemented as a module that is configured to control processor 363 to perform specific functions according to the programming of the module.
[0116] It should be understood that any particular order or hierarchy of steps in the disclosed manufacturing process is an illustration of an exemplary method. Based on design preferences, it is understood that the specific order or hierarchy of steps in the manufacturing process can be rearranged, or only a portion of the steps shown can be performed. Some steps can be performed simultaneously. For example, in some cases, multitasking and parallel processing may be advantageous. Moreover, the separation of various system components in the above-described embodiments should not be understood as requiring such separation in all embodiments, and it should be understood that the described program components and systems can generally be integrated into a software product or packaged into multiple software products.
[0117] The predictive process control apparatus, method and system have been described in detail with particular reference to these illustrated embodiments. However, it is apparent that various modifications and changes may be made within the spirit and scope of the present disclosure as described in the foregoing specification, and these modifications and changes are considered equivalent and are part of the present disclosure.
[0118] Disclosure Statement
[0119] Statement 1. A computer-implemented method comprising:
[0120] The deep learning controller receives a plurality of control values from two or more stations, wherein the control values are generated at the two or more stations deployed in a manufacturing process;
[0121] Predicting, by the deep learning controller, an expected value of an intermediate output or a final output of a manufactured product based on the control value;
[0122] determining, by the deep learning controller, whether the predicted expected value of the artifact meets specifications; and
[0123] If the predicted expected value of the artifact does not meet the specifications, one or more control inputs are generated by the deep learning controller, wherein the one or more control inputs are configured to cause an intermediate output or a final output of the artifact to meet the specifications.
[0124] Statement 2. The computer-implemented method of statement 1, further comprising:
[0125] If a confidence level associated with the one or more control inputs exceeds a predetermined threshold, the one or more control inputs are output to one or more station controllers.
[0126] Statement 3. The computer-implemented method of statement 1 or 2, wherein outputting the one or more control inputs to one or more station controllers further comprises:
[0127] providing a first control input to a first station controller associated with the first station; and
[0128] A second control input is provided to a second station controller associated with a second station, wherein the second station is downstream from the first station in the manufacturing process.
[0129] Statement 4. The computer-implemented method of any of statements 1-3, wherein predicting an intermediate output or a final output of the manufactured product further comprises:
[0130] identifying one or more key impacts associated with the manufacturing process; and
[0131] One or more control inputs associated with at least one of the one or more critical influences are adjusted.
[0132] Statement 5. The computer-implemented method of any one of statements 1-4, further comprising:
[0133] If the predicted expected value of the manufactured product meets the specification, the one or more station controllers are permitted to control the manufacturing process.
[0134] Statement 6. The computer-implemented method of any one of statements 1-5, further comprising:
[0135] The deep learning controller receives a plurality of station values, wherein the station values are associated with the one or more stations, and
[0136] The predicted value of the manufactured product is also predicted based on the station value.
[0137] Statement 7. The computer-implemented method of any of statements 1-6, wherein the plurality of control values comprises one or more of speed, temperature, pressure, vacuum, rotation, current, voltage, resistance, and power.
[0138] Statement 8. A system comprising:
[0139] one or more processors; and
[0140] A non-volatile memory for storing instructions that, when executed by the one or more processors, cause the one or more processors to perform operations including:
[0141] The deep learning controller receives a plurality of control values from two or more stations, wherein the control values are generated at the two or more stations deployed in a manufacturing process;
[0142] Predicting, by the deep learning controller, an expected value of an intermediate output or a final output of a manufactured product based on the control value;
[0143] determining, by the deep learning controller, whether the predicted expected value of the artifact meets specifications; and
[0144] If the predicted expected value of the artifact does not meet the specifications, one or more control inputs are generated by the deep learning controller, wherein the one or more control inputs are configured to cause an intermediate output or a final output of the artifact to meet the specifications.
[0145] Statement 9. The system of statement 8, wherein the processor is further configured to perform operations comprising:
[0146] If a confidence level associated with the one or more control inputs exceeds a predetermined threshold, the one or more control inputs are output to one or more station controllers.
[0147] Statement 10. The system of statement 8 or 9, wherein outputting the one or more control inputs to the one or more station controllers further comprises:
[0148] providing a first control input to a first station controller associated with the first station; and
[0149] A second control input is provided to a second station controller associated with a second station, wherein the second station is downstream from the first station in the manufacturing process.
[0150] Statement 11. The system of any of statements 8-10, wherein predicting the expected value of the manufactured product further comprises:
[0151] identifying one or more key impacts associated with the manufacturing process; and
[0152] One or more control inputs associated with at least one of the one or more critical influences are adjusted.
[0153] Statement 12. The system of any of statements 8-11, further comprising:
[0154] If the predicted expected value of the manufactured product meets the specification, the one or more station controllers are permitted to control the manufacturing process.
[0155] Statement 13. The system of any of statements 8-12, wherein the processor is further configured to perform operations comprising:
[0156] The deep learning controller receives a plurality of station values, wherein the station values are associated with the one or more stations, and
[0157] The predicted value of the manufactured product is also predicted based on the station value.
[0158] Statement 14. The system of any of statements 8-13, wherein the plurality of control values comprises one or more of speed, temperature, pressure, vacuum, rotation, current, voltage, resistance, and power.
[0159] Statement 15. A non-transitory computer-readable storage medium comprising instructions stored therein, which when executed by the one or more processors cause the one or more processors to perform operations comprising:
[0160] The deep learning controller receives a plurality of control values from two or more stations, wherein the control values are generated at the two or more stations deployed in a manufacturing process;
[0161] Predicting, by the deep learning controller, an expected value of an intermediate output or a final output of a manufactured product based on the control value;
[0162] determining, by the deep learning controller, whether the predicted expected value of the artifact meets specifications; and
[0163] If the predicted expected value of the artifact does not meet the specifications, one or more control inputs are generated by the deep learning controller, wherein the one or more control inputs are configured to cause an intermediate output or a final output of the artifact to meet the specifications.
[0164] Statement 16. The non-transitory computer-readable storage medium of statement 15, further comprising instructions configured to cause the one or more processors to perform operations comprising:
[0165] If a confidence level associated with the one or more control inputs exceeds a predetermined threshold, the one or more control inputs are output to one or more station controllers.
[0166] Statement 17. The non-transitory computer-readable storage medium of statement 15 or 16, wherein outputting the one or more control inputs to the one or more station controllers further comprises:
[0167] providing a first control input to a first station controller associated with the first station; and
[0168] A second control input is provided to a second station controller associated with a second station, wherein the second station is downstream from the first station in the manufacturing process.
[0169] Statement 18. The non-transitory computer-readable storage medium of any of statements 15-17, wherein predicting the expected value of the article of manufacture further comprises:
[0170] identifying one or more key influencing factors associated with the manufacturing process; and
[0171] One or more control inputs associated with at least one of the one or more key influencing factors are adjusted.
[0172] Statement 19. The non-transitory computer-readable storage medium of any of statements 15-18, further comprising instructions configured to cause the one or more processors to perform operations comprising:
[0173] If the predicted expected value of the manufactured product meets the specification, the one or more station controllers are permitted to control the manufacturing process.
[0174] Statement 20. The non-transitory computer-readable storage medium of any of statements 15-19, further comprising instructions configured to cause the one or more processors to perform operations comprising:
[0175] The deep learning controller receives a plurality of station values, wherein the station values are associated with the one or more stations, and
[0176] The expected value of the manufactured product is also predicted based on the station value.
Claims
1. A computer-implemented method comprising: The deep learning controller receives a plurality of control values from two or more stations, wherein the plurality of control values are generated at the two or more stations deployed in a manufacturing process; the plurality of control values represent properties of the two or more stations; The deep learning controller predicts, based on the plurality of control values, an expected value of an intermediate output or an expected value of a final output of a manufactured product undergoing the manufacturing process; wherein the manufactured product has undergone processing steps at the two or more stations; determining, by the deep learning controller, whether the predicted expected value of the artifact meets specifications; and If the predicted expected value of the manufactured product does not meet the specifications, one or more control inputs are generated by the deep learning controller for a downstream station among the two or more stations, and the one or more control inputs are configured to make the intermediate output or final output of the manufactured product meet the specifications.
2. The computer-implemented method of claim 1 , further comprising: If a confidence level associated with the one or more control inputs exceeds a predetermined threshold, the one or more control inputs are output to a station controller associated with the downstream station.
3. The computer-implemented method of claim 2, wherein outputting the one or more control inputs to the station controller associated with the downstream station further comprises: providing a first control input to a first station controller associated with a first downstream station; as well as A second control input is provided to a second station controller associated with a second downstream station, wherein the second downstream station is downstream from the first downstream station in the manufacturing process.
4. The computer-implemented method of claim 1 , wherein predicting an expected value of an intermediate output or an expected value of a final output of the artifact further comprises: identifying one or more key influencing factors associated with the manufacturing process; as well as One or more control inputs associated with at least one of the one or more key influencing factors are adjusted.
5. The computer-implemented method of claim 1 , further comprising: If the predicted expected value of the manufactured product meets the specification, one or more station controllers are allowed to control the manufacturing process.
6. The computer-implemented method of claim 1 , further comprising: The deep learning controller receives a plurality of station values, wherein the plurality of station values are associated with the two or more stations, and An expected value of the intermediate output or the expected value of the final output of the product is also predicted based on the multiple station values.
7. The computer-implemented method of claim 1, wherein: The plurality of control values include one or more of speed, temperature, pressure, vacuum, rotation, current, voltage, resistance, and power.
8. A system comprising: one or more processors; as well as A non-volatile memory for storing instructions that, when executed by the one or more processors, cause the one or more processors to perform operations including: The deep learning controller receives a plurality of control values from two or more stations, wherein the plurality of control values are generated at the two or more stations deployed in a manufacturing process; the plurality of control values represent properties of the two or more stations; The deep learning controller predicts, based on the plurality of control values, an expected value of an intermediate output or an expected value of a final output of a manufactured product undergoing the manufacturing process; wherein the manufactured product has undergone processing steps at the two or more stations; determining, by the deep learning controller, whether the predicted expected value of the artifact meets specifications; and If the predicted expected value of the manufactured product does not meet the specifications, one or more control inputs are generated by the deep learning controller for a downstream station among the two or more stations, and the one or more control inputs are configured to make the intermediate output or final output of the manufactured product meet the specifications.
9. The system of claim 8, wherein the processor is further configured to perform operations comprising: If a confidence level associated with the one or more control inputs exceeds a predetermined threshold, the one or more control inputs are output to a station controller associated with the downstream station.
10. The system of claim 9, wherein: Outputting the one or more control inputs to the station controller associated with the downstream station further comprises: providing a first control input to a first station controller associated with a first downstream station; and A second control input is provided to a second station controller associated with a second downstream station, wherein the second downstream station is downstream from the first downstream station in the manufacturing process.
11. The system of claim 8, wherein predicting an expected value of an intermediate output or an expected value of a final output of the manufactured product further comprises: identifying one or more key influencing factors associated with the manufacturing process; as well as One or more control inputs associated with at least one of the one or more key influencing factors are adjusted.
12. The system of claim 8, further comprising: If the predicted expected value of the manufactured product meets the specification, one or more station controllers are allowed to control the manufacturing process.
13. The system of claim 8, wherein the processor is further configured to perform operations comprising: The deep learning controller receives a plurality of station values, wherein the plurality of station values are associated with the two or more stations, and An expected value of an intermediate output or an expected value of a final output of the product is also predicted based on the plurality of station values.
14. The system of claim 8, wherein: The plurality of control values include one or more of speed, temperature, pressure, vacuum, rotation, current, voltage, resistance, and power.
15. A non-transitory computer-readable storage medium comprising instructions stored therein, which, when executed by one or more processors, cause the one or more processors to perform operations comprising: The deep learning controller receives multiple control values from two or more stations, where: The plurality of control values are generated at the two or more stations deployed in a manufacturing process; the plurality of control values represent properties of the two or more stations; The deep learning controller predicts, based on the plurality of control values, an expected value of an intermediate output or an expected value of a final output of a manufactured product undergoing the manufacturing process; wherein the manufactured product has undergone processing steps at the two or more stations; determining, by the deep learning controller, whether the predicted expected value of the artifact meets specifications; and If the predicted expected value of the manufactured product does not meet the specifications, one or more control inputs are generated by the deep learning controller for a downstream station among the two or more stations, and the one or more control inputs are configured to make the intermediate output or final output of the manufactured product meet the specifications.
16. The non-transitory computer-readable storage medium of claim 15, further comprising instructions configured to cause the one or more processors to perform operations comprising: If a confidence level associated with the one or more control inputs exceeds a predetermined threshold, the one or more control inputs are output to a station controller associated with the downstream station.
17. The non-transitory computer readable storage medium of claim 16, wherein: Outputting the one or more control inputs to the station controller associated with the downstream station further comprises: providing a first control input to a first station controller associated with a first downstream station; and A second control input is provided to a second station controller associated with a second downstream station, wherein the second downstream station is downstream from the first downstream station in the manufacturing process.
18. The non-transitory computer-readable storage medium of claim 15, wherein predicting an expected value of an intermediate output or an expected value of a final output of the manufactured product further comprises: identifying one or more key influencing factors associated with the manufacturing process; as well as One or more control inputs associated with at least one of the one or more key influencing factors are adjusted.
19. The non-transitory computer-readable storage medium of claim 15, further comprising instructions configured to cause the one or more processors to perform operations comprising: If the predicted expected value of the manufactured product meets the specification, one or more station controllers are allowed to control the manufacturing process.
20. The non-transitory computer-readable storage medium of claim 15, further comprising instructions configured to cause the one or more processors to perform operations comprising: The deep learning controller receives a plurality of station values, wherein the plurality of station values are associated with the two or more stations, and An expected value of an intermediate output or an expected value of a final output of the product is also predicted based on the plurality of station values.
