Using analytics and algorithms to predict weld quality

Abstract

To provide systems and methods for using analytics and algorithms to predict weld quality.SOLUTION: The systems and methods include a computer having a processor and memory configured to: receive weld parameter data generated during a welding process by a welder to join at least two parts through welding; input the received weld parameter data to a data analytics model to generate at least one predicted weld quality parameter; compare the predicted weld quality parameter with a weld quality parameter threshold; and generate output indicating at least one of the at least one predicted weld quality parameter and a result of the comparison between the at least one predicted weld quality parameter and the weld quality parameter threshold.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] This application claims the benefit of U.S. Provisional Application No. 63 / 224,648, filed Jul. 22, 2021. This application also claims the benefit and priority of Indian Provisional Patent Application 20212107108, filed Apr. 12, 2021. The entire disclosure of each of the above applications is incorporated herein by reference.

[0002] This disclosure relates to systems and methods that use analytics to predict welding quality, and more particularly, to systems and methods that use machine learning, including regression, classification, and / or artificial intelligence models and algorithms, to evaluate and predict the quality of welding.

Background Art

[0003] This section provides background information related to the present disclosure, which is not necessarily prior art.

[0004] Welding devices are used to weld two or more workpieces or parts together. Metal workpieces can be welded together, for example, using an ultrasonic metal welder. Plastic workpieces can be welded together, for example, using an ultrasonic plastic welder, a laser welder, a vibration / friction welder, an infrared welder, a spin welder, a hot plate welder, and / or by using vibration welding with infrared (IR) preheating, also known as clean vibration technology, hot gas welding, radio frequency welding, or thermal staking. <00,00017> Traditionally, to determine weld quality, several workpieces are sampled and tested from a population of welded workpieces at a predetermined sample rate, which can represent up to 100% of the total welded workpieces. Some types of tests, such as peel strength tests and tensile strength tests, have traditionally utilized destructive testing procedures on the workpiece being tested, resulting in good parts being destroyed during the testing process. Depending on the quantity of parts being welded and the sample rate, the costs associated with destructive testing can be substantial. For example, if a factory operates with a production quantity of 30,000 pieces to be welded per day per production line, and a test sample rate of one destructive test per 1,000 pieces, then 30 pieces will be destroyed per day per production line. If the cost of each piece is $30, the cost associated with the destroyed parts would be $900 per day per production line. If a factory has eight production lines, the daily cost due to parts being destroyed would be $7,200, and the resulting annual cost due to parts being destroyed during destructive testing would be $2.6 million. As this example shows, the costs associated with destructive testing can be substantial. [Prior art documents] [Patent Documents]

[0006] [Patent Document 1] U.S. Patent No. 10,722,973 [Patent Document 2] U.S. Patent No. 10,864,608 [Non-patent literature]

[0007] [Non-Patent Document 1] Balz, I., Abi Raad, E., Rosenthal, E., et al. Process monitoring of ultrasonic metal welding battery tabs using external sensor data. Journal of Advanced Joining Processes, Vol. 1 (March 2020). [Non-Patent Document 2] "SMOTE:Synthetic Minority Over-sampling Technique", Journal of Artificial Intelligence Research 16(2002)321-357, published June 2002 [Non-Patent Document 3] "Multivariate Statistical Data Analysis-Principal Component Analysis (PCA)", International Journal of Livestock Research, 7(5), 60-78, (2017) [Non-Patent Document 4] "Feature modeling using polynomial classifiers and stepwise regression," Neurocomputing, Vol. 73, Nos. 10-12, June 2010, pp. 1752-1759. [Non-Patent Document 5] "Generalized Discriminant Analysis algorithm for feature reduction in Cyber ​​Attack Detection", International Journal of Computer Science and Information Security (IJCSIS), Volume 6, Issue 1, 2009 [Non-Patent Document 6] "An overview of gradient descent optimization algorithms," arXiv:1609.94747v2, June 15, 2017. [Overview of the project] [Means for solving the problem]

[0008] This section provides a general overview of the disclosure and does not constitute a full or comprehensive disclosure of all features of the disclosure.

[0009] The system includes a computer having a processor and memory, configured to receive welding parameter data provided and generated during a welding process by a welding machine for joining at least two parts by welding, input the received welding parameter data into a data analysis model for generating at least one predicted welding quality parameter, compare the predicted welding quality parameter with a welding quality parameter threshold, and generate an output indicating at least one of the following: at least one predicted welding quality parameter and at least one of the results of a comparison between the at least one predicted welding quality parameter and the welding quality parameter threshold.

[0010] A method is also provided which includes: a computer having a processor and memory receiving welding parameter data generated during a welding process by a welding machine for joining at least two parts by welding; inputting the received welding parameter data into a data analysis model for generating at least one predicted welding quality parameter; comparing at least one predicted welding quality parameter with a welding quality parameter threshold; and generating an output indicating at least one of at least one predicted welding quality parameter and at least one of the results of a comparison between the at least one predicted welding quality parameter and the welding quality parameter threshold.

[0011] Another system includes a computer having a processor and memory, configured to receive welding parameter data provided and generated during a welding process by a welding machine for joining at least two parts by welding, filter the welding parameter data by removing anomalies in the welding parameter data based on comparing the welding parameter data to at least one threshold, generate an output indicating that for any anomaly in the welding parameter data, the welding parameter data has been rejected, input the filtered welding parameter data into a data analysis model for generating at least one predicted welding quality parameter, compare the predicted welding quality parameter to a welding quality parameter threshold, and generate an output indicating at least one of the following: at least one predicted welding quality parameter and at least one result of a comparison between the at least one predicted welding quality parameter and the welding quality parameter threshold.

[0012] Another system includes a computer having a processor and memory, configured to receive welding parameter data generated during a welding process by a welding machine provided for joining at least two parts by welding, filter the welding parameter data by removing first anomalies in the welding parameter data based on comparing the welding parameter data with high threshold and low threshold, filter the welding parameter data by removing second anomalies in the welding parameter data based on a statistical process control algorithm, generate an output indicating that the welding parameter data has been rejected for the first and second anomalies in the welding parameter data, input the filtered welding parameter data into at least one data analysis model for generating at least one predicted welding quality parameter, compare the predicted welding quality parameter with a welding quality parameter threshold, and generate an output indicating at least one of the results of a comparison between at least one predicted welding quality parameter and the welding quality parameter threshold.

[0013] Another system is provided, which includes a computer having a processor and memory, configured to: receive welding parameter training data including welding parameters and welding quality data associated with the welding parameters; input the received welding parameter training data into multiple data analysis models for generating multiple corresponding predicted welding quality parameters; compare each of the predicted welding quality parameters with known welding quality data; select an individual data analysis model from the multiple data analysis models based on the comparison; receive actual welding parameter data generated during the welding process by a welding machine for joining at least two parts by welding; input the actual welding parameter data into an individual data analysis model for generating at least one predicted welding quality parameter; compare the predicted welding quality parameter with a welding quality parameter threshold; and produce an output indicating at least one of the results of a comparison between at least one predicted welding quality parameter and the welding quality parameter threshold.

[0014] Another method is provided, which includes: a computer having a processor and memory receiving welding parameter training data, including welding parameters and welding quality data associated with the welding parameters; inputting the received welding parameter training data into multiple data analysis models for generating a plurality of corresponding predicted welding quality parameters; comparing each of the predicted welding quality parameters with known welding quality data; selecting individual data analysis models from the plurality of data analysis models based on the comparison; receiving actual welding parameter data generated during the welding process by a welding machine for joining at least two parts by welding; inputting the actual welding parameter data into individual data analysis models for generating at least one predicted welding quality parameter; comparing the predicted welding quality parameter with a welding quality parameter threshold; and generating an output indicating at least one predicted welding quality parameter and at least one of the results of a comparison between the at least one predicted welding quality parameter and the welding quality parameter threshold.

[0015] Further areas of applicability will become apparent from the descriptions provided herein. The descriptions and specific examples in this summary are for illustrative purposes only and are not intended to limit the scope of this disclosure.

[0016] The drawings described herein are for illustrative purposes only, and not to limit the scope of this disclosure, but merely to illustrate selected embodiments. [Brief explanation of the drawing]

[0017] [Figure 1] This is a block diagram of the welding machine and quality monitoring system as disclosed herein. [Figure 2]A block diagram of another welding machine and quality monitoring system according to the present disclosure. [Figure 3] A block diagram of a quality monitoring computer / server according to the present disclosure. [Figure 4] A flow diagram of a quality monitoring method according to the present disclosure. [Figure 5] A block diagram of a quality monitoring module in training mode according to the present disclosure. [Figure 6] A flow diagram of a training process for a quality monitoring system according to the present disclosure. [Figure 7] A block diagram of a quality monitoring module in quality monitoring mode according to the present disclosure. [Figure 8] A flow diagram of a quality monitoring method according to the present disclosure. [Figure 9A] A diagram illustrating a graph of data from a case study of the present disclosure. [Figure 9B] A diagram illustrating a graph of data from a case study of the present disclosure. [Figure 9C] A diagram illustrating a graph of data from a case study of the present disclosure. [Figure 10A] A diagram illustrating a graph of data from a case study of the present disclosure. [Figure 10B] A diagram illustrating a graph of data from a case study of the present disclosure. [Figure 11] A diagram illustrating a graph of data from a case study of the present disclosure. [Figure 12] A block diagram of a quality monitoring system using regression and classification algorithms or models / AI neural network algorithms or models according to the present disclosure. [Figure 13] A block diagram of a quality monitoring system using statistical process control algorithms, as well as regression and classification algorithms or models / AI neural network algorithms or models according to the present disclosure. [Figure 14]This disclosure provides upper and lower limits, statistical process control algorithms, and a block diagram of a quality monitoring system using regression and classification algorithms or models / AI neural network algorithms or models. [Figure 15] This flowchart shows common tasks that are automated by machine learning, as disclosed in this document. [Modes for carrying out the invention]

[0018] The corresponding reference number indicates the corresponding part throughout the various views of the drawing.

[0019] Exemplary embodiments will now be described in more detail with reference to the accompanying drawings.

[0020] This disclosure provides systems and methods that use analytical methods to predict weld quality. For example, this disclosure includes systems and methods that use data analytical methods and machine learning, including regression, classification, and / or artificial intelligence (AI) models and algorithms, to evaluate and predict the quality of metal welds, plastic welds, and / or multi-material welds based on welding parameters monitored or detected during the welding process. While this disclosure provides welding examples, this disclosure is equally applicable to other mechanical joining processes, such as thermal staking, riveting, and / or any other mechanical joining processes, where the quality of the joining process can be evaluated and predicted by this disclosure. Thus, the systems and methods of this disclosure use data analytical methods and machine learning to accurately and efficiently predict one or more metrics of the weld quality of individual workpieces based on monitored or detected welding parameters, potentially reducing or eliminating the resulting destructive testing and discarded production lots of welded workpieces. Therefore, the costs of destructive testing or discarded lots can be mitigated or avoided by using the accurate and efficient testing procedures described herein.

[0021] The systems and methods of this disclosure may utilize data analysis and machine learning algorithms or models, categorized as supervised learning algorithms or models, such as regression and classification algorithms or models. In addition, or alternatively, the systems and methods of this disclosure may utilize data analysis and machine learning algorithms or models, categorized as unsupervised learning algorithms or models, such as AI neural network algorithms or models. For example, the data analysis and machine learning algorithms or models utilized by the systems and methods of this disclosure may include decision tree algorithms or models, random forest algorithms or models, K-nearest neighbors (KNN) algorithms or models, naive Bayes algorithms or models, support vector machine algorithms or models, and / or AI neural network algorithms or models, as will be discussed in more detail below.

[0022] This disclosure includes a quality monitoring system and method for monitoring and receiving data from a welding machine. For example, the monitored and received welding parameter data and / or graphs may include data about current, voltage, power, frequency, time, distance / depth, force, velocity, and / or other data parameters generated or used by the welding machine during the welding process. The welding graph data may include sequential data from the welding process, such as time vs. power, time vs. weld depth, and / or any other appropriate data, which are mapped over time to create the welding graph data. In addition, or alternatively, the quality monitoring system may receive welding parameter data from sensors outside the welding process, such as temperature data from a temperature sensor, laser vibrometer position measurements, stress / strain data, image data of the resulting weld from a camera, vibration data from an impact sensor indicating vibration of an ultrasonic stack, and others. The quality monitoring system and method receive welding parameter data as input and processes the welding parameter data using data analysis theory and machine learning, as discussed above and below, to produce an output indicating and / or corresponding to the quality of the resulting weld. For example, quality monitoring systems and methods can process various inputs and generate and output predicted values ​​indicating the quality of weld strength, such as peel strength, tensile strength / tensile strength, burst pressure, appearance, and others. These predicted values ​​can then be compared to predetermined thresholds to determine whether the welded part has sufficient weld strength / quality. In addition, or alternatively, the quality monitoring system can generate output based on a comparison of predicted values ​​to predetermined thresholds to indicate whether the predicted weld quality metric or measurement is sufficient, based on whether the individual weld followed the expected process and / or achieved any measurement of quality characteristics above or below a predetermined threshold.

[0023] Referring to Figures 1 and 2, example ultrasonic welding systems 10 and 11 are shown in this disclosure, which use data analysis and machine learning to predict weld quality. The ultrasonic welding system 10 shown in Figure 1 is configured to weld at least two metal parts 20, while the ultrasonic welding system 11 shown in Figure 2 is configured to weld at least two plastic parts 30. The at least two parts to be welded may include only two parts to be welded together, or three or more parts to be welded together, such as a bundle of wires, a stack of three or more parts, or any other suitable group of parts to be welded and joined together. The ultrasonic welding systems 10 and 11 each include an ultrasonic power supply unit 18 that provides power to an ultrasonic converter 12, a booster 14, and an ultrasonic horn 16. As is commonly understood by those skilled in the art of ultrasonic welding, the ultrasonic horn 16 is a metal rod that transmits mechanical energy from the ultrasonic converter 12 and booster 14 to the workpiece, i.e., the parts to be welded together. The ultrasonic horn 16 is typically half a wavelength long at the resonant frequency produced by the ultrasonic converter 12. For example, electrical energy from the power supply unit 18 at a frequency of 20-60 kHz is converted into mechanical energy by the ultrasonic converter 12. The mechanical energy converted in the ultrasonic converter 12 is transmitted by the booster 14 and ultrasonic horn 16 to the application, such as a metal part 20 (Figure 1) or a plastic part 30 (Figure 2), which will be welded together. The booster 14 and ultrasonic horn 16 perform the functions of transmitting mechanical energy and changing the mechanical vibration from the ultrasonic converter 12 by a gain factor. The ultrasonic welding machine systems 10 and 11 are controlled by controllers having inputs and outputs coupled to the appropriate components of the ultrasonic welding machine systems 10 and 11, respectively. For example, the power supply unit 18 may include a controller, or the controller may be a separate device from the power supply unit 18. After the welding process, the parts 20 and 30 are joined together by a weld 40.

[0024] The resulting mechanical vibrations on the tip of the horn 16 are the motion that performs the task of welding parts 20 and 30 together. The horn tip can be made of titanium or another high-strength hard material. Parts 20 and 30 to be welded together are placed close to the horn tip, such as by being placed together on an anvil. The horn tip is moved by the actuator 22 to a state in which it is in contact with the upper portions of parts 20 and 30 to be welded together. The ultrasonic horn 16 oscillates in a manner that imparts oscillation to parts 20 and 30 to be welded together. That is, the ultrasonic horn 16 causes oscillating compression / decompression of parts 20 and 30 relative to each other, causing the surfaces of parts 20 and 30 that are in contact with each other at the welding interface to heat up and eventually melt together to form a weld 40.

[0025] Figures 1 and 2 depict ultrasonic welding machine systems 10 and 11 as examples, respectively, but the disclosure is equally applicable to other types of welding machines and workpiece processing devices. For example, in addition to ultrasonic welding machines, the systems and methods of the disclosure that use analytical theories for predicting weld quality are also applicable to laser welding systems, vibration / friction welding systems, infrared welding systems, combined vibration welding with infrared (IR) preheating systems (also referred to as clean vibration technology), spin welding systems, and / or hot plate welding systems. In addition, examples of metal welding machines and plastic welding machines are discussed above with reference to Figures 1 and 2, but the disclosure is equally applicable to other welding machines such as multi-material welding machines, and other mechanical joining processes. In addition, although the disclosure provides welding examples, the disclosure is equally applicable to other mechanical joining processes such as thermal staking, riveting, and / or any other mechanical joining processes, where the quality of the joining process can be evaluated and predicted by the disclosure. Further details relating to ultrasonic welding machines and other workpiece processing devices are contained in U.S. Patent No. 10,722,973, issued on 28 July 2020, entitled "Dynamic Adjustment of Weld Parameter of an Ultrasonic Welder," assigned to the applicant and its assignee, and U.S. Patent No. 10,864,608, issued on 15 December 2020, entitled "Work Piece Processing Device with Servo-Elastic Actuator System with Compliance Elastic Member and Weight Compensation Elastic Member," also assigned to the applicant and its assignee, both of which are incorporated in their entirety by reference.

[0026] As shown in Figures 1 and 2, the ultrasonic welding machine systems 10 and 11 each include a quality monitoring system 50. The quality monitoring system 50 receives various welding parameters related to quality monitoring. For example, the quality monitoring system 50 may receive welding parameters from the ultrasonic power supply unit 18 based on the electrical characteristics of the electrical energy supplied to the ultrasonic converter by the power supply unit 18, including, for example, data indicating the current, voltage, power, and / or frequency of the electrical energy supplied. In addition, the welding parameters may include data relating to the amount of time the electrical energy is supplied by the power supply unit 18, such as the start time, stop time, or duration of the electrical energy supplied for welding. In addition, or alternatively, the quality monitoring system 50 may further receive welding parameters from the actuator 22 based on the characteristics of the motion of the horn 16 by the actuator 22, including, for example, data indicating the distance / depth of the motion of the horn 16, the force of the horn 16 on the parts 20 and 30, and / or the speed or downward speed of the horn 16. In addition, or alternatively, welding parameters from actuator 22 may also include data related to weld collapse and / or weld retention of weld 40. In addition, or alternatively, the quality monitoring system 50 may further receive welding parameters from one or more sensors 42 included within ultrasonic welding machine systems 10, 11. Sensors 42 may include, for example, one or more temperature sensors, laser vibrometers, position sensors, stress / strain sensors, impact sensors, and / or cameras. A temperature sensor may measure and provide temperature data to the quality monitoring system 50. A laser vibrometer may measure and provide vibration data of the ultrasonic stack, or vibration data of the vibration welding machine in a vibration welding machine embodiment. A position sensor may provide position data of the horn 16 and / or the parts 20, 30 being welded. A stress / strain sensor may provide stress / strain data of the parts 20, 30 being welded. An impact sensor may provide vibration data of the ultrasonic stack. A camera may provide image data of the parts 20, 30 being welded.

[0027] The quality monitoring system 50, the power supply unit 18, the actuator 22, and / or the sensor 42, if any, each include sufficient communication interfaces for communicating such data via a wired or wireless communication connection with the quality monitoring system 50.

[0028] In some embodiments, the quality monitoring system 50 may include a quality monitoring computer located in the same location as the ultrasonic welding machine systems 10, 11. In such cases, the ultrasonic power supply unit 18, actuators 22, and / or sensors 42, if any, can communicate various welding parameters and data to the quality monitoring computer for processing according to the present disclosure, and the quality monitoring computer may be connected to an output device such as a computer monitor and / or a display or touchscreen device for providing output to the user of the quality monitoring computer. In other embodiments, the quality monitoring computer may be integrated with the welding machine. For example, the quality monitoring computer may be embedded in the ultrasonic power supply unit 18 and connected to an output device for providing output to the user. In other embodiments, the quality monitoring system may be integrated with the end user's manufacturing enterprise system (MES). In other embodiments, the quality monitoring system 50 may include a quality monitoring client computer located in the same location as the ultrasonic welding machine systems 10, 11 and a quality monitoring server computer located remotely at a different location from the ultrasonic welding machine systems 10, 11. In such cases, the quality monitoring client computer can receive various welding parameters and data from the ultrasonic power supply unit 18, actuator 22, and / or sensors 42, if any, and communicate these various welding parameters and data to the quality monitoring server for processing according to this disclosure. For example, the quality monitoring computer may be implemented in a cloud environment remotely from the location or facility of the ultrasonic welding machine system 10, 11, such as within a cloud environment using Microsoft Azure cloud services, Amazon Web Services, or Google cloud computing services. As a further example, the quality monitoring computer / system may be implemented using the Plantweb™ digital ecosystem available from Emerson Electric Co.

[0029] Referring to Figure 3, the quality monitoring computer / server 300 of the quality monitoring system 50 is shown. As mentioned above, in some embodiments, the quality monitoring computer / server 300 can function as a standalone computer and may be located in the same location or facility as the ultrasonic welding machine systems 10, 11, in which case it can directly receive various welding parameters from the ultrasonic power supply unit 18, actuators 22, and / or sensors 42, if any. In other embodiments, the quality monitoring computer / server 300 can function as a server and may be located remotely from the ultrasonic welding machine systems 10, 11, for example, in a cloud environment or within a manufacturing enterprise system (MES). In such cases, a quality monitoring client can receive various welding parameters and data from the ultrasonic power supply unit 18, actuators 22, and / or sensors 42, if any, and communicate these various welding parameters and data to the quality monitoring computer / server 300, which is functioning as a server.

[0030] The quality monitoring computer / server 300 may include a central processing unit (CPU) 302, memory 304, a network interface 306, volatile storage 308, and an input / output interface 310. Memory 304 stores executable code for the quality monitoring computer / server 300, for application 312, and for the operating system (OS) 314. Memory 304 also stores executable quality monitoring code 316 for the quality monitoring module, training code 318 for the training module, data analysis code 320 for the data analysis module, and quality prediction code 322 for the quality prediction module, which are discussed in more detail below. The OS 314 handles coordinating the input / output of hardware resources and executing the binary executable file of the quality monitoring computer / server 300, including the quality monitoring code 316, training code 318, data analysis code 320, and quality prediction code 322. The quality monitoring computer / server 300 may be a physical device with dedicated hardware, integrated with the welding machine, and / or embedded within the welding machine. Alternatively, the quality monitoring computer / server 300 may be implemented as a virtual machine sharing a portion of the dedicated hardware resources.

[0031] Referring to Figure 4, an overview of the methodology for systems and methods using data analysis and machine learning to evaluate and predict weld quality, as described herein, is illustrated. First, as will be discussed in more detail below, training data 400 is used in a learning or training process 402 to develop and train a data analysis model 404, such as a regression model, a classification model, and / or an AI neural network model. The data analysis model 404 may include, for example, a random forest model, a decision tree model, a support vector regression model, and / or a gradient boosting regression model such as Catboost or SGBoost. The training data 400 may include, for example, various welding parameters or graphs stored in relation to known or actual welding quality parameters. In the training process, the welding parameters are used to train the data analysis model 404 to generate one or more predicted welding quality parameters based on the input welding parameters. The predicted welding quality parameters are then compared to the actual welding quality parameters for the associated input welding parameters, and the data analysis model 404 may then be adjusted, modified, and / or tuned based on the difference between the predicted and actual welding quality parameters. The learning or training process 402 then continues for the next set of input welding parameters from the training data 400. The learning or training process 402 continues and is repeated until the data analysis model 404 has the ability to robustly and accurately generate predicted welding quality parameters that are the same as the actual welding quality parameters for a given set of input welding parameters, or within a predetermined tolerance of those actual welding quality parameters. For example, the learning or training process 402 may continue and be repeated until the accuracy of the quality prediction by the data analysis model 404 is greater than a predetermined accuracy threshold, such as 95% accuracy. Alternatively, the learning or training process 402 may continue and be repeated until all of the training data 400 has been utilized.Once the training process is complete, the data analysis model 404 can be used in the prediction process 406 with respect to actual live welding parameter data to predict welding quality parameters for actual welding performed by actual welding machine systems such as ultrasonic welding machine systems 10, 11.

[0032] In the example in Figure 4, the prediction process 406 is used to generate weld quality parameters corresponding to the predicted peel strength of the weld. As illustrated in Figure 4, in the prediction process 406, the predicted peel strength is compared to a peel threshold that is 12 Newtons, or 12N, in this example. As a result of the comparison, and / or the predicted weld quality parameters, such as the predicted peel strength, are output and communicated to a quality control process, which then further evaluates and / or disposes of, replaces, or repairs products having welds with predicted weld quality parameters, such as predicted peel strength, that are below a predetermined threshold.

[0033] Referring to Figure 5, a block diagram showing the quality monitoring module 500 in training mode is shown. The quality monitoring module 500 includes a training module 502, a data analysis module 504, and a quality prediction module 506. The quality monitoring module 500, training module 502, data analysis module 504, and quality prediction module 506 can be implemented by a quality monitoring computer / server 300 that runs quality monitoring codes 316, training code 318, data analysis code 320, and quality prediction code 322, which are stored in memory 304 as shown in Figure 3 and discussed above. As shown in Figure 5, the quality monitoring module 500 receives training data 400 from a computer-readable storage medium accessible to the quality monitoring module 500 and the quality monitoring computer / server 300. As discussed above, the training data 400 may include various welding parameters, which are stored in relation to known or actual welding quality parameters. In training mode, the training module 502 of the quality monitoring module 500 acquires stored welding parameters corresponding to actual welds, which have known actual welding quality parameters, and inputs these stored welding parameters into the data analysis module 504, which then utilizes the data analysis model 404 to generate one or more predicted welding quality parameters based on the input welding parameters. The training module 502 then compares the predicted welding quality parameters with the actual welding quality parameters for the associated input welding parameters, which are included in the training data and associated with the input welding parameters. Based on the comparison, the training module 502 can instruct the data analysis module 504 to adjust, modify, and tune the data analysis model 404 based on the difference between the predicted welding quality parameters and the actual welding quality parameters. The training module 502 then continues with the next set of input welding parameters from the training data 400.The quality monitoring module 500 and the training module 502 continue to operate in this manner in training mode until the data analysis module 504 robustly and accurately generates predicted welding quality parameters that are the same as, or within a predetermined tolerance of, the actual welding quality parameters for a given set of input welding parameters from the training data, or until the quality monitoring module 500 and the training module 502 run out of additional entries in the set of training data 400. For example, the quality monitoring module 50 and the training module 502 can operate in training mode until the accuracy of the quality prediction by the data analysis module 504 exceeds a predetermined accuracy threshold, such as 95% accuracy. Alternatively, the quality monitoring module 500 and the training module 502 can continue to operate in training mode until all of the training data 400 has been used. At that point, the quality monitoring module 500 may be switched from training mode to quality prediction mode and may be used with respect to actual live welding parameter data to predict welding quality parameters for actual welding performed by actual welding machine systems such as ultrasonic welding machine systems 10, 11.

[0034] Referring to Figure 6, Method 600 for the training process is shown, starting in 602. Method 600 can be performed by the Quality Monitoring Computer / Server 300, as well as, for example, the Training Module 502 and Data Analysis Module 504 of the Quality Monitoring Module 500. In 604, Training Module 502 obtains welding parameter data and actual welding quality parameter data from training data 400. In 606, Training Module 502 inputs the welding parameter data into Data Analysis Module 504, which in turn inputs the welding parameter data into Data Analysis Model 404 to generate predicted welding quality parameters. In 608, Training Module 502 compares the predicted welding quality parameters with the actual welding quality parameters, and in 610, Training Module 502 instructs Data Analysis Module 504 to adjust Data Analysis Model 404 based on the comparison. In 612, Training Module 502 determines whether any additional entries are available in the training data 400 for processing. If, at step 612, there are no additional entries available in the training data 400, method 600 terminates at step 616. If, at step 612, additional entries are available in the training data 400, the training module proceeds to step 614. At step 614, the training module 502 determines whether to continue operating in training mode by comparing the current accuracy of the predicted welding quality parameters with an accuracy threshold. For example, the training module 502 may compare the current accuracy with a predetermined accuracy threshold, such as 95% accuracy. If the data analysis module 504 generates predicted welding quality parameters with an accuracy greater than the predetermined accuracy threshold, the training module 502 proceeds to step 616 and terminates the training process. If, at step 614, the current accuracy is less than the predetermined accuracy threshold, the training module 502 loops back to step 604 and retrieves the next set of data from the training data 400.

[0035] Referring to Figure 7, a block diagram showing the quality monitoring module 500 in quality monitoring mode is shown. In quality monitoring mode, the quality prediction module 506 receives welding parameter input data from actual welding machine systems such as ultrasonic welding machine systems 10, 11. As discussed in detail above, the welding parameter input data may include welding parameter data received from the ultrasonic power supply unit 18, from the actuator 22, and / or from the sensor 42, if any. The quality prediction module 506 then inputs the received welding parameter input data to the data analysis module 504, which inputs the received welding parameter input data into the data analysis model 404 to generate predicted welding quality parameters for the welding based on the input welding parameter data. The data analysis module 504 returns the predicted welding quality parameters to the quality prediction module 506, which then outputs the predicted welding quality parameters. The predicted welding quality parameters may be compared to predetermined threshold quality in a quality control process performed by the quality monitoring module 500 and / or by a separate quality control process or device. If the predicted welding quality parameters are below the predetermined threshold quality, the weld and / or associated product may be further evaluated and / or disposed of, replaced, or repaired.

[0036] In quality monitoring mode, the quality monitoring module 500 can utilize a fixed algorithm, which fixes the data analysis module 504 and the data analysis model 404 at the end of the training process and prevents further updates, adjustments, modifications, and / or tuning during live operation in quality monitoring mode. Alternatively, the quality monitoring module 500 can utilize an active learning algorithm, which uses actual welding to be tested on-site, and the test data to compare the actual welding quality parameters with the predicted welding quality parameters output by the quality monitoring module 500. In such a case, the quality monitoring module can then update, adjust, modify, and / or tune the data analysis model 404 based on the comparison results so that the accuracy of the data analysis module 504 and the quality prediction module 506 can be improved during live operation and adapted to the actual operation and welding conditions on-site at the facility.

[0037] Referring to Figure 8, Method 800 for the quality monitoring process is shown, beginning in 802. Method 800 may be performed by a quality monitoring computer / server 300, as well as, for example, a quality prediction module 506 and a data analysis module 504 of a quality monitoring module 500. In 804, the quality prediction module 506 receives welding parameter input data from an actual welding machine system, such as an ultrasonic welding machine system 10, 11. In 806, the quality prediction module 506 inputs the welding parameter data into the data analysis module 504, which in turn inputs the welding parameter data into a data analysis model 404 for generating predicted welding quality parameters. In 808, the quality prediction module 506 receives the predicted welding quality parameters from the data analysis module 504 and compares the predicted welding quality parameters with a predetermined quality threshold. In 810, the quality prediction module 506 generates an output indicating the predicted weld quality parameters and / or the result of a comparison between the predicted weld quality parameters and a predetermined quality threshold. For example, if the predicted weld quality parameters are below a predetermined threshold quality, the weld and / or associated product may be subjected to further evaluation and / or disposed of, replaced, or repaired. Alternatively, in some embodiments, the comparison in 808 may be omitted, and the quality prediction module 506 may simply output the predicted weld quality parameters in 810. In 811, the quality monitoring process ends.

[0038] Data Analysis Theory Algorithms The quality monitoring systems and methods of the present disclosure utilize data analysis and machine learning, including regression, classification, and / or AI models and algorithms, to predict or estimate the welding quality for a given weld based on welding parameters determined, generated, and / or detected during the welding process. As discussed above, the systems and methods of the present disclosure may utilize supervised learning algorithms or models, such as regression and classification algorithms or models, and / or unsupervised learning algorithms or models, such as AI neural network algorithms or models. For example, the data analysis and machine learning algorithms or models utilized by the systems and methods of the present disclosure may include decision tree algorithms or models, random forest algorithms or models, K-nearest neighbors (KNN) algorithms or models, naive Bayes algorithms or models, support vector machine algorithms or models, and / or AI neural network algorithms or models. In some embodiments, the quality monitoring systems and methods of the present disclosure may include gradient boosting regression algorithms for decision trees, such as gradient boosting regression algorithms for decision trees referred to as CatBoost and / or SGBoost. The CatBoost gradient boosting regression algorithm includes features such as ordered boosting to overcome overfitting, native handling of categorical features, and the use of oblivious or symmetric trees for faster execution. The XGBoost gradient boosting regression algorithm includes features such as clever penalization of trees, proportional shrinking of leaf nodes, Newton boosting, extra randomization parameters, and implementation on single, distributed systems, and out-of-core computation.In some embodiments, the quality monitoring systems and methods of the present disclosure may include a random forest regression algorithm, also referred to as a random decision forest, which is an ensemble machine learning algorithm. The random forest regression algorithm includes an ensemble learning method that operates by building a large number of decision trees during the training process and outputting a class that is a mode of a class, or a prediction of the mean / average of the individual trees.

[0039] In other embodiments, quality monitoring systems and methods of the present disclosure may include the use of artificial neural networks to predict or estimate the welding quality for a given weld based on welding parameters determined, generated, and / or detected during the welding process.

[0040] While specific data analysis algorithms are discussed above, any other machine learning (supervised / structured or unstructured / unsupervised) algorithms, including functionality for predicting or estimating parameters such as weld quality based on input parameters determined, generated, and / or detected during processes such as welding, may be used by this disclosure.

[0041] In addition, matrix profiling anomalies can be used to explore welding graph data for anomalies, as will be discussed in more detail below.

[0042] Welding quality parameters The quality monitoring systems and methods of the present disclosure predict or estimate the weld quality of a weld based on input welding parameters that are determined, generated, and / or detected during the welding process. As discussed in the examples above, the quality monitoring systems of the present disclosure can generate predicted values ​​indicating the quality of the weld strength, such as peel strength and / or tensile strength. While specific weld quality parameters are discussed in the examples above, any methodology for evaluating weld quality may be used. For example, the quality monitoring systems and methods of the present disclosure can predict or estimate weld quality by predicting or estimating any of the following weld quality parameters:

[0043] - Gas leak detection to measure the integrity of welded seals.

[0044] - Testing embedded electronics by applying energy to them and measuring the result. For example, a taillight may be lit to ensure that the welding process did not damage the electronics inside.

[0045] - Tensile strength. A destructive method for measuring weld strength by pulling a part apart.

[0046] - Peel strength. A destructive method for separating a weld. It is more frequently used in wire splicing. Peel strength is often different from tensile strength (tensile strength).

[0047] - Breaking or bursting strength. A destructive method in which a container is pressurized until it bursts. This is a metric for weld strength.

[0048] - Critical length measurement. This can be measured, for example, by a die, a tool for checking fit, a ruler, or a laser measuring device. (The measurement may be of the entire part or of some dimension of a weld, such as weld depth).

[0049] - Thermal detection. Welding often results in higher, localized temperatures within the material. Thermocouples, IR cameras, IR thermometers, or laser thermometer beams / guns may be used to measure the temperature of the weld area.

[0050] - Automated Optical Inspection (AOI) / Video Image Processing. Customers take digital photos or videos of welds for record keeping and quality control. This is becoming more widespread in the medical industry where traceability is more important.

[0051] - Appearance. Visual inspection by humans or machines is often used to check weld quality. Inspectors look for bubbles, strands, excess weld material (also known as "flash"), weld gaps (insufficient weld), cracks, stress lines, burn marks, and other visual features. In metal welding, customers particularly look for discoloration and signs of oxidation. Microscopes and magnification may be used.

[0052] - Measurement methods such as optical 3-D surface scanning. Suppliers such as Gom, Zeiss, and Nikon supply machines that use variable focus or lasers to determine surface dimensions. 3D scans of parts can show how those parts look before or after welding, compared to an ideal or specified part.

[0053] - Measure residual internal stress. The welding process can create internal partial stresses, which are often relieved using thermal annealing (in the furnace).

[0054] - Drop test. Components are dropped according to standard test methods to measure damage.

[0055] - Electrical resistance or conductivity. Used in wire and spot metal welding.

[0056] - Optical absorption. Plastics may be transparent or opaque after welding, and this can be observed or measured.

[0057] - X-ray tomography. X-rays can be used to image the internal structure of a part or assembly in order to look for abnormalities.

[0058] - Surface profiling or surface roughness. Measurement of the surface topology or roughness of a part, either near or away from a weld. Many types of devices exist that can measure this metric, such as profilers, scanning tunneling microscopes (STMs), linear variable differential transformers, interferometer profilers, and scanning probes.

[0059] - Shear strength. A destructive method of applying shear force to welding.

[0060] - Resistance to crushing, compression, or deformation. A destructive test that measures the compressive load on a sample to deform, cleave, shatter, or collapse an assembly.

[0061] - Creep test. Used to determine the deformation of a component under stress over time.

[0062] - Thermal testing. High temperature, low temperature, or thermal cycling to measure changes in the main properties over time.

[0063] - Humidity or salt spray testing. Exposing customer products to harsh environmental conditions.

[0064] - Young's modulus test: The rate at which strain changes in response to stress on a material.

[0065] - Peel strength test. Similar to the peel test, but applied to thin film applications. A "climbing drum" may be used.

[0066] - Hardness testing. Measuring the durometer of a material or component. Testing equipment such as a Rockwell hardness tester may be used.

[0067] - Hermetic properties (also known as airtightness or watertightness). Welds and containers can be tested for leaks.

[0068] - Cross-section cutting. A destructive method for cutting open welds or parts.

[0069] - Chemical analysis. Welded parts and assemblies can be tested for chemical composition.

[0070] - Gas chromatography. Used to measure gas leakage to determine the chemical composition.

[0071] - Particulate matter measurement. Measurement against dust or weight (parts per million, or particles per volume) - more likely in cleanrooms, air ducts, and microfluidics.

[0072] - Fatigue. Repeated motion used to evaluate the lifespan or cycle of a part.

[0073] - Density. A measurement of the weight per unit volume of the components of an assembly.

[0074] - X-ray diffraction. X-rays are emitted onto a material to measure the profile of how the X-rays diffract after striking elemental particles in the sample.

[0075] - Contamination. Inspecting welds for the introduction of unusual substances, foreign materials, or excessive materials (for example, when welding toothpaste tubes, the inspection process was to look for toothpaste on the outside of the tube).

[0076] - Resonance. Testing to find the resonant frequency of assembly components (we often ask customers to redesign components because those components vibrate at a resonant frequency close to our welding machine frequency).

[0077] Closed-loop correction As discussed in detail above, the quality monitoring systems and methods of the present disclosure utilize data analysis theory and machine learning algorithms and models to predict or estimate the welding quality for a given weld based on welding parameters determined, generated, and / or detected during the welding process. In addition to predicting or estimating welding quality, the quality monitoring systems and methods of the present disclosure may include systems and methods for performing closed-loop corrections to the welding process in order to adjust the operating characteristics of the welding process for subsequent welds. For example, to adjust and enhance welding quality, the quality monitoring systems and methods of the present disclosure may include closed-loop corrections to adjust the motion of the horn 16 and the force applied by the horn 16 to adjust the electrical energy supplied to the ultrasonic converter 12 by the power supply unit 18, and / or to adjust any other controllable aspects of the welding process.

[0078] Case study #1 The dataset was generated from 200 welds for a Branson GSX-E1 welding machine operated in energy mode according to a set recipe for plastic i-beam parts. Measured variables included maximum force, weld collapse, welding time, peak power, and others. Post-weld tensile strength was performed and recorded as shown in Figures 9A-9C, 10A, and 10B. The following models were run on the data to predict tensile strength values ​​and quantitative measurements of weld quality: - Decision tree algorithm - Random Forest Algorithm - K-Nearest Neighbors - Naive Bayes algorithm - Support Vector Machine (SVM) algorithm - Artificial neural network algorithms The results were as follows:

[0079] [Table 1]

[0080] Case study #2 Another dataset was generated from 500 welds from a Branson GMX-W1 wire splicer. The following models were run on the data to predict and estimate weld quality: random forest, decision tree, support vector regression (SVM), XGBoost, Catboost, and neural network. The results were as follows:

[0081] [Table 2]

[0082] These case studies conclude that methodologies using machine learning on conventional power supply data can develop algorithms that accurately predict both metal welded and plastic joined products. This conclusion contradicts current cutting-edge thinking in the field, as summarized in the paper: Balz, I., Abi Raad, E., Rosenthal, E. et al. Process monitoring of ultrasonic metal welding battery tabs using external sensor data. Journal of Advanced Joining Processes, Vol. 1 (March 2020).

[0083] Case Study #3: Anomalies in Matrix Profiling According to this disclosure, a matrix profiling anomaly method may be used to explore welding graph data for anomalies, as will be discussed in more detail below.

[0084] A method called matrix profiling anomaly was used to explore welding graph data for welding anomalies. A visual representation is shown in Figure 11. In this method, the parameter-versus-time graph (in this case, weld depth-versus-time and power-versus-time) is divided into smaller sets of data, such as 50 data points or 50ms intervals. These sets are then compared to known anomalies. Peaks in the matrix profile graph indicate significant changes in performance.

[0085] Referring to Figure 12, block diagrams of example ultrasonic welding machine systems 10 and 11 are shown, which include a quality monitoring system 50, an ultrasonic power supply unit 18, an actuator 22, and one or more sensors 42, if any, as described above. The ultrasonic welding machine device 900 may include, for example, a converter 12, a booster 14, and a horn 16, as described above with reference to Figures 1 and 2 (not shown in Figure 12). As described above, the quality monitoring system 50 may include a data analysis module 504 that uses data analysis and machine learning algorithms or models, including supervised learning algorithms or models such as regression and classification algorithms or models, or unsupervised learning algorithms or models such as AI neural network algorithms or models. The algorithms used by the quality monitoring system 50, for example, regression and classification algorithms or models / AI neural network algorithms or models, are shown in 904 in Figure 12. As described above, the quality monitoring system 50 may receive welding parameter inputs, such as the electrical properties of the electrical energy supplied to the ultrasonic converter by the power supply unit 18, including data indicating the current, voltage, power, and / or frequency of the electrical energy supplied. As further described above, the quality monitoring system 50 may also receive welding parameter inputs, such as data relating to the amount of time the electrical energy is supplied by the power supply unit 18, such as the start time, stop time, or duration of the electrical energy supplied for welding. In addition, or alternatively, the quality monitoring system 50 may also receive welding parameters from the actuator 22, based on the characteristics of the motion of the horn 16 by the actuator 22, including data indicating the distance / depth of the motion of the horn 16, the force of the horn 16 on the parts 20, 30, and / or the velocity or downward velocity of the horn 16. In addition, or alternatively, the welding parameters from the actuator 22 may also include data relating to the weld collapse and / or weld retention of the weld 40.In addition, or alternatively, the quality monitoring system 50 may also receive welding parameters from one or more sensors 42 included within the ultrasonic welding machine systems 10, 11. The sensors 42 may include, for example, one or more temperature sensors, laser vibrometers, position sensors, stress / strain sensors, impact sensors, and / or cameras. The temperature sensor may measure and provide temperature data to the quality monitoring system 50. The laser vibrometer may measure and provide vibration data of the ultrasonic stack, or vibration data of the vibration welding machine in a vibration welding machine embodiment. The position sensor may provide position data of the horn 16 and / or the parts 20, 30 being welded. The stress / strain sensor may provide stress / strain data of the parts 20, 30 being welded. The impact sensor may provide vibration data of the ultrasonic stack. The camera may provide image data of the parts 20, 30 being welded. As described above, the quality monitoring system 50 utilizes data analysis and machine learning algorithms or models, as shown in Figure 12, 904, to receive welding parameter data, generate predicted welding quality parameters, and / or generate the results of comparing the predicted welding quality parameters with predetermined thresholds. The generated predicted welding quality parameters and / or the results of comparing the predicted welding quality parameters with predetermined thresholds are shown in Figure 12, 906. Based on the comparison, for example, the quality monitoring system 50 can classify the corresponding welds as either good or questionable. For example, if the welding quality parameters correspond to the predicted peel strength of a weld, the quality monitoring system 50 can classify the corresponding welds as good if the predicted peel strength is greater than the predetermined threshold, and as questionable if the predicted peel strength is less than the predetermined threshold.The quality monitoring system 50 can then output the generated predicted welding quality parameters and / or the results of a comparison of the predicted welding quality parameters to the human-machine interface (HMI) display 908, i.e., a classification of either good or questionable. For example, the HMI display 908 can display actual predicted welding quality parameters, such as the actual predicted tensile strength of the weld, and / or the results of a comparison of the welding quality parameters with predetermined thresholds.

[0086] A quality monitoring system 50, shown at 904 in Figure 12, which utilizes a single data analysis theory and machine learning algorithm or model, can determine predicted welding quality parameters based on received welding parameter inputs. However, in some cases, the quality monitoring system 50 can be confused by welding parameter input data that contains significant anomalies. In other words, in some cases, a quality monitoring system 50 utilizing a single data analysis theory and machine learning algorithm or model may have difficulty predicting the welding quality of a weld when the welding parameter input data deviates too far from normal or average welding parameter input data.

[0087] For this reason, in another embodiment shown with reference to Figure 13, the quality monitoring system 50' may include an additional data processing layer of welding parameter input data before the welding parameter data is input to a data analysis and machine learning algorithm or model, i.e., a regression and classification algorithm or model / AI neural network algorithm or model, as shown at 904 in Figure 13. For example, the quality monitoring system 50' may utilize a statistical process control (SPC) algorithm, as shown at 910 in Figure 13, to process the welding parameter input data before it is input to the regression and classification algorithm or model / AI neural network algorithm or model at 904, and to remove any welds that correspond to significant anomalies in the welding parameter input data. For example, the SPC algorithm 910 of the quality monitoring system 50' may analyze each data value of the welding parameter input data, compare that data value to the SPC mean of the welding parameter input data, and reject any welds that correspond to the welding parameter input data that are farther from the SPC mean by a predetermined number of standard deviations. As shown in Figure 13, welds rejected as suspicious based on SPC data analysis are indicated at 912 in Figure 13. Outputs indicating rejected / suspicious welds are output to the HMI display 908. If the data values ​​of the welding parameter input data are within a predetermined number of standard deviations from the SPC mean, the welding parameter data is input to a regression and classification algorithm or model / AI neural network algorithm or model, shown at 904 in Figure 13, for processing as discussed in detail above with respect to Figure 12. As described above, the predicted welding quality parameters and / or the results of comparing the predicted welding quality parameters with predetermined thresholds may be displayed on the HMI display 908.

[0088] Thus, the accuracy of the quality monitoring system 50' can be improved by using a number of algorithms, including the SPC algorithm 910 for analyzing the welding parameter input data and removing significant anomalies before the welding parameter data is input to the regression and classification algorithm or the model / AI neural network algorithm or model 904. In this way, the welding parameter data is preprocessed or filtered to remove significant anomalies before being processed by the regression and classification algorithm or the model / AI neural network algorithm or model 904. For example, welds that produce welding parameter input data with significant anomalies may be the result of: an excessive number of layers of metal foil, too few layers of metal foil, material samples with oil on them that thereby reduce friction, use of the wrong metal, welding without the proper end piece or cover, welding in the wrong area, welding a single sample twice, oxidized material, horn tip not properly aligned with the fixture, use of the wrong horn or anvil, and / or any other problem that produces significant anomalies in the welding parameter input data.

[0089] Figure 13 illustrates the use of two layers of data processing, namely the SPC algorithm 910 and the regression and classification algorithm or model / AI neural network algorithm or model 904, but additional layers of data processing may also be used to further increase the accuracy of the system. For example, referring to Figure 14, another embodiment of the quality monitoring system 50” that utilizes three or more algorithms is shown. The quality monitoring system 50” is similar to the quality monitoring system 50’ to the quality monitoring system 50’ that utilizes upper and lower limit algorithms 914 before the SPC algorithm 910. The upper and lower limit algorithms 914 may be used, for example, to compare data values ​​of welding parameter input data with high and low thresholds. Any welds with corresponding welding parameter input data that are outside the upper and lower limits, i.e., higher than the upper threshold and lower than the lower threshold, may be rejected as suspicious in 912. Any welds with corresponding welding parameter input data that are within the high and low thresholds may be passed to the SPC algorithm for processing similar to that described above, referring to the quality monitoring system 50’ shown in Figure 12. The upper and lower limits may be predetermined thresholds and / or user-adjustable thresholds. For example, the upper and lower limits may be initialized by default threshold settings that can be adjusted by the user of the quality monitoring system 50”.

[0090] Three algorithms are shown in Figure 13, but any number (N) of algorithms can be used. For example, additional layers of regression and classification algorithms or models, and / or AI neural network algorithms or models, may be included between the SPC algorithm 910 and the regression and classification algorithm or model / AI neural network algorithm or model 904. Furthermore, the regression and classification algorithm or model / AI neural network algorithm or model shown in 904 can be decomposed into a number of layers. For example, one layer of data processing may include processing using a regression and classification algorithm or model, while a separate layer may include processing using an AI neural network algorithm or model. For example, a quality monitoring system may utilize four layers of data processing, including: an upper and lower bound algorithm layer; an SPC algorithm layer; a regression and classification layer utilizing a decision tree algorithm, random forest algorithm, KNN algorithm, naive Bayes algorithm, or others; and an AI neural network algorithm layer.

[0091] Thus, the use of multiple layers of data processing algorithms can more effectively remove significant anomalies in welding parameter input data and more accurately predict welding quality. Therefore, the use of multiple layers of data processing algorithms makes the system more robust and less susceptible to noise and anomalous data.

[0092] As in the embodiments described above, the HMI display 908 can display outputs indicating predicted weld quality parameters and / or the results of a comparison of the predicted weld quality parameters with predetermined thresholds. The HMI display 908 can further display outputs indicating that individual welds have been rejected as questionable based on individual processing algorithms, such as the upper and lower limit algorithms 914 or the SPC algorithm 910.

[0093] Case Study #4: End-to-End Example According to this disclosure, another set of approximately 1,000 records of process data was generated in the laboratory and used to train the model. In this case study, the ultrasonic process emanating from a welding machine was inherently highly nonlinear, with negative correlations between fields and very few failure cases. The training data was preprocessed for missing attributes, attribute values, noise or outliers, and duplicates before modeling.

[0094] In addition, the Synthetic Minority Oversampling Technique (SMOTE) was applied to overcome data imbalance. SMOTE is described in "SMOTE: Synthetic Minority Over-sampling Technique," Journal of Artificial Intelligence Research 16(2002)321-357, published in June 2002.

[0095] The training data included the following columns: cycle count, tensile force, energy, trigger pressure, pressure, amplitude, time, power, pre-height, and height.

[0096] Modeling & Training Process As part of the modeling exercise, the feature importance for each feature (column) was calculated using: Principal Component Analysis (PCA), a dimensionality reduction machine learning technique, as described in "Multivariate Statistical Data Analysis - Principal Component Analysis (PCA)," International Journal of Livestock Research, 7(5), 60-78, (2017); polynomial features, as described in "Feature modeling using polynomial classifiers and stepwise regression," Neurocomputing, Vol. 73, Nos. 10-12, June 2010, pp. 1752-1759; and Generalized Discriminant Analysis (GDA), as described in "Generalized Discriminant Analysis algorithm for feature reduction in Cyber ​​Attack Detection," International Journal of Computer Science and Information Security (IJCSIS), Vol. 6, No. 1, 2009.

[0097] The training data was divided into 70% training data and 30% test data, and five algorithms for nonlinear data were trained using their respective hyperparameters. Model weights were initialized randomly, and stochastic gradient descent (SGD), as described in "An overview of gradient descent optimization algorithms," arXiv:1609.94747v2, June 15, 2017, was used to adjust the weights as part of the training.

[0098] All models had the following outputs, as shown here: features "power (N)", "welding time (seconds)", "height (mm)", and "pre-height (mm)", as well as "tensile force (N)".

[0099] [Table 3]

[0100] Model Evaluation The trained models were tested and evaluated on 30% validation data, and also on a synthetic version of real-world data that preserved almost all the statistical properties of the original data. Random Forest and CatBoost ultimately produced the best performance. The results for model evaluation are shown in the table below:

[0101] [Table 4]

[0102] The model was trained on a dataset with 700 rows. Predictions were based on a dataset with 300 rows.

[0103] Model prediction The best performing model was selected to be deployed as part of a deployable analysis engine code.

[0104] During live welding sessions, a process data flow was configured to the model so that it would receive process data for every weld, and the model would then use that process data to predict the resulting potential "tensile forces."

[0105] [Table 5]

[0106] The predicted tensile force is then compared to the minimum threshold tensile force that determines the weld quality. If the predicted tensile force is below the threshold, the weld job is classified as questionable. Otherwise, the weld job is classified as good.

[0107] Case Study #5: Automated Machine Learning According to this disclosure, automated machine learning features enable users of the welding analysis engine to train their own models using their data on-premises, avoiding the need to send production data out for offline modeling and training.

[0108] This feature allows the welding analysis engine to automatically evaluate and select the best algorithm and / or stack of algorithms, along with optimal hyperparameter values, for a given training dataset. Thus, this automated machine learning feature automates the selection, composition, and parameterization of machine learning models, leveraging recent advances in genetic programming, stochastic Bayesian optimization, and ensemble construction.

[0109] The system automates some of the common tasks required to build and evaluate a good predictive model or ensemble, along with optimal hyperparameter values, for a given dataset. For example, as shown in Figure 15, the system automates the following tasks, which include: data cleaning 1502, feature reduction and selection 1504, model selection 1506, hyperparameter tuning 1508, and model validation 1510. As a result, the automated system builds the best pipeline of actions to be performed on a good predictive mode / ensemble model. The pipeline is frequently updated as the system attempts to find better-performing algorithms and tuning values ​​by trying to fit many machine learning algorithms, along with their own parameter values ​​and numerous preprocessing steps, to a given dataset.

[0110] The system uses a genetic programming algorithm to find the best scoring pipeline by comparing cross-validation scores, i.e., accuracy, across all generations. The system explores a wide range of estimators, feature selectors, feature constructors, algorithms, models, and parameters to find the sequence of operators that minimizes the error in model predictions. The best scores across numerous generations are shown below (best pipeline: LinearSVR(input_matrix,C=5.0,dual=False,epsilon=1.0,loss=squared_spsilon_insensitive,tol=0.1)):

[0111] [Table 6]

[0112] The table below contains two predicted tensile force series: (1) predictions from a fine-tuned model; and (2) predictions from automated machine learning (AutoML). The data corresponds to an automotive battery welding application, and the resulting test accuracy was 97.8%.

[0113] [Table 7] TIFF2022162537000009.tif106170

[0114] Thus, the systems and methods of the present disclosure include using different algorithms and models with respect to test data, evaluating and selecting an algorithm or model that most accurately predicts weld quality, and then using the selected algorithm or model to predict weld quality using live weld parameter data.

[0115] In addition, as a quality monitoring system is used in the field over time, the welding machine hardware and other system components may suffer process drift and may not perform the same way as when the quality monitoring system was first installed. Therefore, this disclosure includes updating the quality monitoring system to tune and check the selected algorithms or models used by the quality monitoring system in order to address process drift resulting from the system's use in the field over time. For example, the quality monitoring system may be used with respect to a new set of test data, and the algorithms or models used by the internal parameters used by the quality monitoring system may be retuned and recalibrated in the field to address process drift and to more accurately predict weld quality. In other words, the data analysis model may be recalibrated after the welding machine has been operated in the field by utilizing tuning procedures based on actual welding data that correlate welding parameter data with welding quality data.

[0116] The foregoing description is merely illustrative and is not intended to limit the Disclosure, its uses, or applications. The broad teachings of the Disclosure can be realized in various forms. Therefore, although the Disclosure includes individual examples, the true scope of the Disclosure should not be limited in this way, for other modifications will become apparent from the drawings, this specification, and the following claims. It should be understood that one or more steps in the method may be performed in a different order (or simultaneously) without altering the principles of the Disclosure. Furthermore, although each embodiment is described above to have certain features, any one or more of those features described for any embodiment of the Disclosure may be realized in any other embodiment and / or in combination with the features of any of those embodiments, even if such combination is not explicitly described. In other words, the embodiments described are not mutually exclusive, and the substitution of one or more embodiments with one another remains within the scope of the Disclosure.

[0117] The spatial and functional relationships between elements (for example, between modules) are described using a variety of terms, including “connected,” “engaged,” “interfaced,” and “joined.” Unless explicitly described as “direct,” when a relationship between a first element and a second element is described in the above disclosure, that relationship includes both a direct relationship in which there are no other intervening elements present between the first and second elements, and an indirect relationship in which one or more intervening elements are present between the first and second elements (either spatially or functionally).

[0118] As used herein, the phrase, at least one of A, B, and C, should be understood to mean (A OR B OR C) when using the non-exclusive logic OR. For example, the phrase, at least one of A, B, and C, should be understood to include: (i) A alone; (ii) B alone; (iii) C alone; (iv) A and B together; (v) A and C together; (vi) B and C together; (vii) any one of A, B, and C together. The phrase, at least one of A, B, and C, should not be understood to mean "at least one of A, at least one of B, and at least one of C".

[0119] In diagrams, the direction of arrows, indicated by their arrowheads, generally indicates the flow of information (such as data or commands) of interest to the example. For example, if elements A and B exchange various types of information, and the information transmitted from element A to element B is relevant to the example, the arrow may point from element A to element B. This one-way arrow does not imply that there is no other information transmitted from element B to element A. Furthermore, in response to information sent from element A to element B, element B may send a request for information or an acknowledgment of receipt of information to element A. A term or subset does not necessarily have to be a subset of the exact meaning. In other words, a first subset of a first set may have the same extent as the first set (is equal to the first set).

[0120] In this application, which includes the definitions below, the terms “module” or “controller” may be replaced with the term “circuit.” The term “module” refers to, or may include, a portion of, processor hardware (shared, dedicated, or group) that executes code, and memory hardware (shared, dedicated, or group) that stores the code executed by the processor hardware.

[0121] A module may contain one or more interface circuits. In some examples, the interface circuit may implement a wired or wireless interface that connects to a local area network (LAN) or a wireless personal area network (WPAN). Examples of LANs include IEEE standard 802.11-2016 (also known as the WIFI wireless networking standard) and IEEE standard 802.3-2015 (also known as the ETHERNET wired networking standard). Examples of WPANs include IEEE standard 802.15.4 (including the ZIGBEE standard from the ZigBee Alliance) and the Bluetooth wireless networking standard from the Bluetooth Special Interest Group (SIG) (including core specification versions 3.0, 4.0, 4.1, 4.2, 5.0, and 5.1 from the Bluetooth SIG).

[0122] Modules may communicate with other modules using interface circuits. While modules may be described in this disclosure as communicating directly and logically with other modules, in various implementations modules may actually communicate through a communication system. A communication system may include physical and / or virtual networking devices such as hubs, switches, routers, and gateways. In some implementations, a communication system may be connected to or traverse a wide area network (WAN), such as the Internet. For example, a communication system may include a number of LANs connected to each other over the Internet or point-to-point rental lines using technologies including Multiprotocol Label Switching (MPLS) and Virtual Private Networks (VPNs).

[0123] In various implementations, the functionality of a module may be distributed among multiple modules connected by a communication system. For example, multiple modules may perform the same functionality distributed by a load balancing system. In a further example, the functionality of a module may be divided between server (and even remote or cloud) modules and client (or user) modules. For example, a client module may include a native or web application that runs on a client device and communicates over a network with the server module.

[0124] The terms and code used above may include software, firmware, and / or microcode, and may refer to programs, routines, functions, classes, data structures, and / or objects. Shared processor hardware encompasses a single microprocessor that runs some or all of the code from a number of modules. Group processor hardware, in combination with additional microprocessors, encompasses a microprocessor that runs some or all of the code from one or more modules. References to a number of microprocessors include a number of microprocessors on a discrete die, a number of microprocessors on a single die, a number of cores on a single microprocessor, a number of threads on a single microprocessor, or a combination of the above.

[0125] Shared memory hardware encompasses a single memory device that stores some or all of the code from multiple modules. Group memory hardware, in combination with other memory devices, encompasses a memory device that stores some or all of the code from one or more modules.

[0126] The term memory hardware is a subset of the term computer-readable media. As used herein, the term computer-readable media does not encompass transient electrical or electromagnetic signals that propagate through a medium (such as on carrier waves); therefore, the term computer-readable media is considered tangible and non-transient. Non-exclusive examples of non-transient computer-readable media include non-volatile memory devices (such as flash memory devices, erasable programmable read-only memory devices, or mask read-only memory devices), volatile memory devices (such as static random-access memory devices or dynamic random-access memory devices), magnetic storage media (such as analog or digital magnetic tapes, or hard disk drives), and optical storage media (such as CDs, DVDs, or Blu-ray discs).

[0127] The apparatus and methods described in this application may be partially or completely implemented by a special-purpose computer, which is made by configuring a general-purpose computer to perform one or more individual functions that are embodied in the form of computer programs. The functional blocks and flowchart elements described above serve as software specifications that can be translated into computer programs through the routine work of a skilled technician or programmer.

[0128] A computer program includes processor-executable instructions stored on at least one non-temporary computer-readable medium. A computer program may also include, or rely on, stored data. A computer program may encompass a basic input / output system (BIOS) that interacts with the hardware of a special-purpose computer, device drivers that interact with individual devices of a special-purpose computer, one or more operating systems, user applications, background services, background applications, and others.

[0129] A computer program may include: (i) descriptive text to be parsed, such as HTML (Hypertext Markup Language), XML (Extensible Markup Language), or JSON (JavaScript Object Notation); (ii) assembly code; (iii) object code generated from the source code by a compiler; (iv) source code for execution by an interpreter; (v) source code for compilation and execution by a just-in-time compiler; and other elements. As a mere example, source code may be written using syntax from languages ​​including C, C++, C#, Objective-C, Swift, Haskell, Go, SQL, R, Lisp, Java(R), Fortran, Perl, Pascal, Curl, OCaml, JavaScript(R), HTML5 (Hypertext Markup Language, 5th Revision), Ada, ASP (Active Server Pages), PHP (PHP: Hypertext Preprocessor), Scala, Eiffel, Smalltalk, Erlang, Ruby, Flash(R), Visual Basic(R), Lua, MATLAB, SIMULINK, and Python(R).

[0130] Exemplary embodiments are provided so as to ensure that this disclosure is thorough and fully conveys its scope to those skilled in the art. Numerous specific details, such as examples of specific components, devices, and methods, are discussed to provide a thorough understanding of the embodiments of this disclosure. It will be apparent to those skilled in the art that specific details are not required to be used, that the exemplary embodiments may be embodied in many different forms, and that neither should be construed as limiting the scope of this disclosure. In some exemplary embodiments, well-known processes, well-known device structures, and well-known techniques are not described in detail.

[0131] The terminology used herein is for the purpose of describing individual exemplary embodiments and is not intended to be limiting. As used herein, the singular forms “a,” “an,” and “the” may also include the plural unless the context otherwise explicitly indicates. The terms “equip,” “possess,” “include,” and “have” are inclusive and therefore determine the existence of the features, integers, steps, actions, elements, and / or components described, but do not exclude the existence or addition of one or more other features, integers, steps, actions, elements, components, and / or groups thereof. The method steps, processes, and actions described herein should not be understood as necessarily requiring their performance in the individual order discussed or illustrated unless explicitly identified as the order of performance. It should also be understood that additional or alternative steps may be used. [Explanation of Symbols]

[0132] 10, 11 Ultrasonic welding machine system 12 ultrasonic converters, converters 14 Booster 16. Ultrasonic horn, horn 18 Ultrasonic power supply section, power supply section 20 Metal parts, components 22 Actuators 30 plastic parts, components 40 Welding 42 sensors 50, 50', 50” Quality Monitoring System 300 Quality Monitoring Computers / Servers 302 Central Processing Unit (CPU) 304 memory 306 Network Interfaces 308 Volatile Storage 310 Input / Output Interfaces 312 Applications 314 Operating Systems (OS) 316 Executable quality monitoring codes, quality monitoring codes 318 Training Code 320 Data Analysis Theory Code 322 Quality Prediction Code 400 training data 402 Learning or training process 404 Data Analysis Theory Models 406 Prediction Process 500 Quality Monitoring Modules 502 Training Modules 504 Data Analysis Theory Module 506 Quality Prediction Module 600, 800 ways 900 Ultrasonic Welding Machine Devices 904 Regression and classification algorithms or models / AI neural network algorithms or models 908 Human-Machine Interface (HMI) Display 910 SPC Algorithm 914 Upper and Lower Bound Algorithms 1502 Data Cleaning 1504 Feature Reduction and Selection 1506 Model Selection 1508 Hyperparameter Tuning 1510 Model Validation

Claims

1. a processor, and when executed by the processor, the processor receiving welding parameter data generated during a welding process by a welding machine for joining at least two parts by welding, the welding parameter data including welding graph data having sequential welding parameter values ​​over a time period of the welding process; inputting the received weld parameter data into a data analytics model for generating at least one predicted weld quality parameter; analyzing the weld graph data using matrix profiling to compare the predicted weld quality parameters to weld quality parameter thresholds and to identify anomalies in the weld graph data; generating at least one predicted weld quality parameter based on the comparing and analyzing, and an output indicative of at least one of a result of the comparison between the at least one predicted weld quality parameter and a weld quality parameter threshold value; and a memory storing instructions for configuring the computer to: A system comprising:

2. The system of claim 1 , wherein the data analytics model is trained during a training procedure based on actual welding data that correlates welding parameter data with weld quality data.

3. 2. The system of claim 1, wherein the data analytics model is based on at least one of a supervised learning algorithm, an unsupervised learning algorithm, a regression algorithm, a classification algorithm, an artificial intelligence neural network algorithm, a decision tree algorithm, a random forest algorithm, a K-nearest neighbor algorithm, a Naive Bayes algorithm, a support vector machine algorithm, and a gradient boosting regression algorithm.

4. The system of claim 1 , wherein the at least one predicted weld quality parameter represents at least one of peel strength, tensile strength, burst pressure, and optical appearance of the weld.

5. 10. The system of claim 1, wherein the welding parameter data generated during the welding process includes data generated by an ultrasonic power supply of the welder corresponding to electrical energy supplied by the ultrasonic power supply of the welder to an ultrasonic converter of the welder, the data including at least one of current data, voltage data, power data, and frequency data.

6. 10. The system of claim 1, wherein the welding parameter data generated during the welding process includes data generated by an actuator that moves a horn of a welder including at least one of horn distance / depth of movement, force, and horn speed / down speed.

7. 10. The system of claim 1, wherein the welding parameter data generated during the welding process includes data generated by at least one sensor, the at least one sensor including at least one of a temperature sensor, a laser vibrometer, a position measurement sensor, a stress / strain sensor, an impact sensor, and a camera.

8. The system of claim 1 , wherein the at least two components include a bundle of wires.

9. 10. The system of claim 1, wherein the welder is one of an ultrasonic welder, a laser welder, a vibration / friction welder, an infrared welder, a combination vibration welder with an infrared preheat system, a spin welder, and a hot plate welder.

10. 10. The system of claim 1, wherein the data analytics model is configured to be recalibrated after the welder is operated in the field by utilizing a tuning procedure based on actual welding data that correlates welding parameter data with weld quality data.

11. receiving, by a computer having a processor and a memory, welding parameter data generated during a welding process by a welding machine for joining at least two components by welding, the welding parameter data including welding graph data having sequential welding parameter values ​​over a time period of the welding process; inputting, by the computer, the received welding parameter data into a data analytics model for generating at least one predicted weld quality parameter; comparing, by the computer, the at least one predicted weld quality parameter to a weld quality parameter threshold and analyzing, by the computer, the weld graph data using matrix profiling to identify anomalies in the weld graph data; generating, by the computer, an output indicative of at least one of the at least one predicted weld quality parameter and a result of the comparison between the at least one predicted weld quality parameter and a weld quality parameter threshold based on the comparing and analyzing; A method comprising:

12. The method of claim 11 , further comprising training a data analytics model during a training procedure based on actual welding data that correlates welding parameter data with weld quality data.

13. 12. The method of claim 11, wherein the data analytics model comprises at least one of a supervised learning algorithm, an unsupervised learning algorithm, a regression algorithm, a classification algorithm, an artificial intelligence neural network algorithm, a decision tree algorithm, a random forest algorithm, a K-nearest neighbor algorithm, a Naive Bayes algorithm, a support vector machine algorithm, and a gradient boosting regression algorithm.

14. The method system of claim 11 , wherein the at least one predicted weld quality parameter represents at least one of a peel strength, a tensile strength / tensile strength, a burst pressure, and an optical appearance of the weld.

15. 12. The method of claim 11, wherein the welding parameter data generated during the welding process includes data generated by an ultrasonic power supply of the welder corresponding to electrical energy supplied by the ultrasonic power supply of the welder to an ultrasonic converter of the welder, the data including at least one of current data, voltage data, power data, and frequency data.

16. 12. The method of claim 11, wherein the welding parameter data generated during the welding process includes data generated by an actuator moving a horn of a welder including at least one of horn distance / depth of movement, force, and horn speed / down speed.

17. 12. The method of claim 11, wherein the welding parameter data generated during the welding process includes data generated by at least one sensor, the at least one sensor including at least one of a temperature sensor, a laser vibrometer, a position measurement sensor, a stress / strain sensor, an impact sensor, and a camera.

18. receiving welding parameter data generated during a welding process by a welding machine for joining at least two parts by welding; filtering the welding parameter data by removing anomalies in the welding parameter data based on comparing the welding parameter data to at least one threshold; generating an output for any anomaly in the welding parameter data indicating that the welding parameter data has been rejected; inputting the filtered weld parameter data into a data analytics model for generating at least one predicted weld quality parameter; Comparing the predicted weld quality parameter to a weld quality parameter threshold; generating an output indicative of at least one of the at least one predicted weld quality parameter and a result of a comparison between the at least one predicted weld quality parameter and a weld quality parameter threshold value; A computer having a processor and a memory configured to A system comprising:

19. 20. The system of claim 18, wherein removing anomalies in the welding parameter data based on comparing the welding parameter data to at least one threshold comprises utilizing a statistical process control algorithm.

20. 20. The system of claim 19, wherein the statistical process control algorithm compares at least one data value of the welding parameter data to a mean value corresponding to the at least one data value, and classifies the at least one data value as anomalous in response to the at least one data value being more than a predetermined number of standard deviations away from the mean value.

21. 20. The system of claim 18, wherein removing anomalies in the welding parameter data based on comparing the welding parameter data to at least one threshold comprises comparing the welding parameter data to a high threshold and a low threshold and classifying the welding parameter data as anomalous in response to the welding parameter data being greater than the high threshold or less than the low threshold.

22. 22. The system of claim 21, wherein the high and low thresholds are user definable.

23. 20. The system of claim 18, wherein the data analytics model is based on at least one of a supervised learning algorithm, an unsupervised learning algorithm, a regression algorithm, a classification algorithm, an artificial intelligence neural network algorithm, a decision tree algorithm, a random forest algorithm, a K-nearest neighbor algorithm, a Naive Bayes algorithm, a support vector machine algorithm, and a gradient boosting regression algorithm.

24. receiving welding parameter data generated during a welding process by a welding machine for joining at least two parts by welding; filtering the welding parameter data by removing a first anomaly in the welding parameter data based on comparing the welding parameter data to a high threshold and a low threshold; filtering the welding parameter data by removing a second anomaly in the welding parameter data based on a statistical process control algorithm; generating an output indicating that the welding parameter data has been rejected in response to the first and second anomalies in the welding parameter data; inputting the filtered weld parameter data into at least one data analytics model for generating at least one predicted weld quality parameter; Comparing the predicted weld quality parameter to a weld quality parameter threshold; generating an output indicative of at least one of the at least one predicted weld quality parameter and a result of a comparison between the at least one predicted weld quality parameter and a weld quality parameter threshold value; A computer having a processor and a memory configured to A system comprising:

25. 25. The system of claim 24, wherein the statistical process control algorithm compares at least one data value of the welding parameter data to a mean value corresponding to the at least one data value, and classifies the at least one data value as anomalous in response to the at least one data value being more than a predetermined number of standard deviations away from the mean value.

26. 25. The system of claim 24, wherein the at least one data analytics model comprises at least one of a supervised learning algorithm, an unsupervised learning algorithm, a regression algorithm, a classification algorithm, an artificial intelligence neural network algorithm, a decision tree algorithm, a random forest algorithm, a K-nearest neighbor algorithm, a Naive Bayes algorithm, a support vector machine algorithm, and a gradient boosting regression algorithm.

27. 25. The system of claim 24, wherein the at least one data analytics model comprises at least two of a supervised learning algorithm, an unsupervised learning algorithm, a regression algorithm, a classification algorithm, an artificial intelligence neural network algorithm, a decision tree algorithm, a random forest algorithm, a K-nearest neighbor algorithm, a Naive Bayes algorithm, a support vector machine algorithm, and a gradient boosting regression algorithm.

28. 25. The system of claim 24, wherein the at least one data analytics model includes an artificial intelligence neural network algorithm and at least one of a regression algorithm, a classification algorithm, a decision tree algorithm, a random forest algorithm, a K-nearest neighbor algorithm, a Naive Bayes algorithm, a support vector machine algorithm, and a gradient boosting regression algorithm.

29. receiving welding parameter training data including welding parameters and weld quality data associated with the welding parameters; inputting the received welding parameter training data into a plurality of data analytics models for generating a plurality of corresponding predicted weld quality parameters; comparing each of the predicted weld quality parameters to known weld quality data; Selecting an individual data analytics model from the plurality of data analytics models based on the comparison; receiving actual weld parameter data generated during a welding process by a welding machine for joining at least two parts by welding, the weld parameter data including weld graph data having sequential weld parameter values ​​over a time period of the welding process; inputting the actual weld parameter data into a separate data analytics model for generating at least one predicted weld quality parameter; analyzing the weld graph data using matrix profiling to compare the predicted weld quality parameters to weld quality parameter thresholds and to identify anomalies in the weld graph data; generating at least one predicted weld quality parameter based on the comparing and analyzing, and an output indicative of at least one of a result of the comparison between the at least one predicted weld quality parameter and a weld quality parameter threshold value; A computer having a processor and a memory configured to A system comprising:

30. 30. The system of claim 29, wherein each of the plurality of data analytics models is based on at least one of a supervised learning algorithm, an unsupervised learning algorithm, a regression algorithm, a classification algorithm, an artificial intelligence neural network algorithm, a decision tree algorithm, a random forest algorithm, a K-nearest neighbor algorithm, a Naive Bayes algorithm, a support vector machine algorithm, and a gradient boosting regression algorithm.

31. 30. The system of claim 29, wherein the at least one predicted weld quality parameter represents at least one of peel strength, tensile strength, burst pressure, and optical appearance of the weld.

32. receiving, by a computer having a processor and a memory, welding parameter training data including welding parameters and weld quality data associated with the welding parameters; inputting, by the computer, the received welding parameter training data into a plurality of data analytics models for generating a plurality of corresponding predicted weld quality parameters; comparing, by the computer, each of the predicted weld quality parameters to known weld quality data; selecting, by the computer, a respective data analytics model from the plurality of data analytics models based on the comparison; receiving, by a computer, actual weld parameter data generated during a welding process by a welding machine for joining at least two components by welding, the weld parameter data including weld graph data having sequential weld parameter values ​​over a time period of the welding process; inputting, by a computer, the actual weld parameter data into a separate data analytics model for generating at least one predicted weld quality parameter; comparing, by the computer, the predicted weld quality parameter to a weld quality parameter threshold value and analyzing, by the computer, the weld graph data using matrix profiling to identify anomalies in the weld graph data; generating, by the computer, an output indicative of at least one of the at least one predicted weld quality parameter and a result of the comparison between the at least one predicted weld quality parameter and a weld quality parameter threshold based on the comparing and analyzing; A method comprising:

33. 33. The method of claim 32, wherein each of the plurality of data analytics models is based on at least one of a supervised learning algorithm, an unsupervised learning algorithm, a regression algorithm, a classification algorithm, an artificial intelligence neural network algorithm, a decision tree algorithm, a random forest algorithm, a K-nearest neighbor algorithm, a Naive Bayes algorithm, a support vector machine algorithm, and a gradient boosting regression algorithm.

34. 33. The method of claim 32, wherein the at least one predicted weld quality parameter represents at least one of peel strength, tensile strength, burst pressure, and optical appearance of the weld.

Images