WO2019103772A1

WO2019103772A1 - Detection and root cause analysis of welding defects

Info

Publication number: WO2019103772A1
Application number: PCT/US2018/047022
Authority: WO
Inventors: Janani VENUGOPALAN; Songtao Xia; Krzysztof CHALUPKA
Original assignee: Siemens Aktiengesellschaft; Siemens Corporation
Priority date: 2017-11-22
Filing date: 2018-08-20
Publication date: 2019-05-31

Abstract

Methods for weld analysis and corresponding systems and computer-readable mediums. A method (300) includes simulating (302) a welding operation (102) by the weld analysis system (400). The method includes analyzing (304) a physical weld (112) corresponding to the simulated welding operation. The method includes detecting and predicting (306) welding defects based on the simulated welding operation and the analysis of the physical weld. The method includes performing a root cause analysis (124, 308), based on the detected and predicted welding defects, to identify at least one root cause of the welding defects. The method includes producing a corrective action (128, 310) based on the identified at least one root cause to remove the welding defects from subsequent physical welds.

Description

DETECTION AND ROOT CAUSE ANALYSIS OF WELDING DEFECTS

CROSS-REFERENCE TO OTHER APPLICATION

[0001] This application claims the benefit of the filing data of United States Provisional Patent Application 62/589,589, filed November 22, 2017, which is hereby incorporated by reference.

TECHNICAL FIELD

[0002] The present disclosure is directed, in general, systems and methods for machine detection of physical defects in part welds.

BACKGROUND OF THE DISCLOSURE

[0003] Automated robotic welding devices can produce welding defects when faulty sensors, programs or parameters are used, and if those defects are not detected, they may be repeated many times. Improved systems are desirable.

SUMMARY OF THE DISCLOSURE

[0004] Various disclosed embodiments include methods for weld analysis and corresponding systems and computer-readable mediums. A method includes simulating a welding operation by a weld analysis system. The method includes analyzing a physical weld corresponding to the simulated welding operation. The method includes detecting and predicting welding defects based on the simulated welding operation and the analysis of the physical weld. The method includes performing a root cause analysis, based on the detected and predicted welding defects, to identify at least one root cause of the welding defects. The method includes producing a corrective action based on the identified at least one root cause to remove the welding defects from subsequent physical welds. In some cases, the method can further include performing physical welding operations according to the corrective action.

[0005] In some embodiments, the simulated welding operation uses one or more of a robotic path, part parameters, weld parameters, and part movements. In some embodiments, the analysis of the physical weld includes one or more of analyzing camera images, analyzing ultrasound images, analyzing sensor inputs, and analyzing weld parameters. In some embodiments, detecting and predicting welding defects is performed by building one or more deep neural networks from results of the simulated welding operation and results of the physical weld analysis. In some embodiments, the root cause analysis includes building causal graphical models using predicted defects based on the simulated welding operation and unpredicted, detected defects from the physical weld analysis. In some embodiments, the corrective action includes corrected motion paths for robotic welders.

[0006] The foregoing has outlined rather broadly the features and technical advantages of the present disclosure so that those skilled in the art may better understand the detailed description that follows. Additional features and advantages of the disclosure will be described hereinafter that form the subject of the claims. Those skilled in the art will appreciate that they may readily use the conception and the specific embodiment disclosed as a basis for modifying or designing other structures for carrying out the same purposes of the present disclosure. Those skilled in the art will also realize that such equivalent constructions do not depart from the spirit and scope of the disclosure in its broadest form.

[0007] Before undertaking the DETAILED DESCRIPTION below, it may be advantageous to set forth definitions of certain words or phrases used throughout this patent document: the terms“include” and“comprise,” as well as derivatives thereof, mean inclusion without limitation; the term“or” is inclusive, meaning and/or; the phrases “associated with” and“associated therewith,” as well as derivatives thereof, may mean to include, be included within, interconnect with, contain, be contained within, connect to or with, couple to or with, be communicable with, cooperate with, interleave, juxtapose, be proximate to, be bound to or with, have, have a property of, or the like; and the term “controller” means any device, system or part thereof that controls at least one operation, whether such a device is implemented in hardware, firmware, software or some combination of at least two of the same. It should be noted that the functionality associated with any particular controller may be centralized or distributed, whether locally or remotely. Definitions for certain words and phrases are provided throughout this patent document, and those of ordinary skill in the art will understand that such definitions apply in many, if not most, instances to prior as well as future uses of such defined words and phrases. While some terms may include a wide variety of embodiments, the appended claims may expressly limit these terms to specific embodiments.

BRIEF DESCRIPTION OF THE DRAWINGS

[0008] For a more complete understanding of the present disclosure, and the advantages thereof, reference is now made to the following descriptions taken in conjunction with the accompanying drawings, wherein like numbers designate like objects, and in which:

[0009] Figure 1 illustrates a block diagram of processes performed by a weld analysis system as disclosed herein in accordance with disclosed embodiments;

[0010] Figure 2 illustrates a process for using controlled experiments to identify faults in a real-time analysis network in accordance with disclosed embodiments;

[0011] Figure 3 illustrates a flowchart of a process in accordance with disclosed embodiments; and

[0012] Figure 4 illustrates a block diagram of a data processing system in which an embodiment can be implemented.

DETAILED DESCRIPTION

[0013] The Figures discussed below, and the various embodiments used to describe the principles of the present disclosure in this patent document are by way of illustration only and should not be construed in any way to limit the scope of the disclosure. Those skilled in the art will understand that the principles of the present disclosure may be implemented in any suitably arranged device. The numerous innovative teachings of the present application will be described with reference to exemplary non-limiting embodiments.

[0014] Spot welding is widely used in the automotive industry to assemble vehicle body parts and can be almost completely automated using robots, and similar techniques are made used in other industries. One major challenge for automated robotic welding devices is the detection, correction, and prevention of welding defects. Unless the cause of the defects is known, the robot will continue the weld process using the same parameters, which often results in an increase in the defects and the repetition of defects. In an industrial welding system, each type of defect occurs due to the interaction between a diverse set of factors, which makes the root-cause analysis a challenge. Fault analysis often requires the systematic elimination of factors to arrive at the true cause. This process is both time-taking and resource-intensive because current practices require manual, human intervention for defect identification, root-cause analysis, and defect correction. This makes the quality and cost of welding for large automotive companies suboptimal.

[0015] Disclosed embodiments incorporate machine-learning-based process for automatic low-cost detection and correction of the welding defects using root-cause analysis, without requiring human intervention. Based on the root causes identified, corrective action is performed and the robot can be used again for re-welding the defective areas.

[0016] Disclosed embodiments include processes for robot weld simulation and weld defect data collection, processes for deep-learning for weld defect detection, processes for causal inference for root cause analysis, and processes for corrective path generation for spot welding robots.

[0017] Disclosed embodiments improved the performance of robotic devices resulting in faster turn arounds for industrial spot welding. Disclosed embodiment can analyze defect data to ensure high weld quality using techniques from perception and sensing. In addition, disclosed processes can be applied to other manufacturing processes including machining defect detection and correction.

[0018] Robotic spot welders can rapidly produce a large number of welds. However, to ensure quality, ultrasonic and destructive testing of assemblies must be periodically conducted. Such manual product sampling tests are time-consuming and the results reflect only the quality achieved in that particular process.

[0019] Adaptive control systems can be used to monitor current, voltage, resistance, etc. during automated welding and compare their values in real-time against predetermined process response curves to assure the required quality has been produced. However, even these advanced methods suffer from the fact that they rely on predetermined values and responses of the relevant variables. The quality and reliability clearly depends on the stored“master curves” or other predetermined data.

[0020] Some defect detection mechanisms, including machine learning-based approaches, focus on the detection of multiple types of welding defects. Research on the defect correction has been largely based on the robotic path planning and re-welding of defective locations. The challenge with such an approach is that the root causes of the defects are not detected and corrected. This can lead to an increase in the defects and repetition of defects.

[0021] Current root-cause analysis techniques for welding defects use statistical techniques which focus on single root causes. Such an approach often requires the systematic elimination of factors to arrive at the true cause, and often requires a human to detect defects and takes corrective action to compensate for the defects created during robotic arm welding. This is time and resource intensive. [0022] Disclosed processes are particularly applicable to the automotive industry, power and gas, and large-scale manufacturing segments which require fast weld times for a large number of parts.

[0023] Figure 1 illustrates a block diagram 100 of processes performed by a weld analysis system as disclosed herein in accordance with disclosed embodiments.

[0024] As illustrated in Fig. 1, a system collects data from simulation experiments 102 and controlled experiments 112 for the training of the defect detection and root cause analysis models. Simulation experiments 102 can include such elements as robotic path 104, part parameters 106, weld parameters 108, and part movement 110. Simulation experiments can be carried out using computer-aided design (CAD), computer-aided manufacturing (CAM), or computer-aided engineering (CAE) simulation software, such as the NX® CAD/CAM/CAE software products by Siemens Product Lifecycle Management Software Inc. (Plano, Texas).

[0025] Controlled experiments 112 are performed by analysis of physical welds, both real-time (as welds are being created) and offline. The controlled experiments can include such elements as camera/ultrasound images 114 of the welds and sensor readings 116 of the welds, including weld parameters.

[0026] The results of both simulation experiments 102 and controlled experiments 112 are then used for defect prediction and detection 118. Defect prediction and detection 118 can include, in this example, a deep neural network to predict defects 120, which can also predict defect types, and in particular can receive feature data and other output from simulation experiments 102. Defect prediction and detection 118 can also include, in this example, a deep neural network to detect defects at runtime 122, which can receive images and other output from controlled experiments 112.

[0027] Defect prediction and detection 118 can produce output including both predicted defects and unpredicted defects, for example those that were detected by the deep neural network to detect defects at runtime 122 from the controlled experiments 112. The output of defect prediction and detection 118 can be sent to root cause analysis 124. [0028] Root cause analysis 124 analyzes the output of defect prediction and detection 118 to determine root causes of the welding defects. This can be accomplished, for example, using causal graphical models 126 to evaluate the root causes. Root cause analysis 124 can also provide feedback, such as the determined root causes or other data, back to the deep neural network to predict defects 120, so that the neural networks can be refined using deep-learning techniques.

[0029] Root cause analysis 124 can send the determined root causes for the system to determine the appropriate corrective action 128. The system can do so, for example, using path planning simulations for corrective actions 130.

[0030] In a process as illustrated in Fig. 1, the deep-models for the defect prediction (which can include auto-encoders and/or feed-forward networks), defect detection (which can include convolutional neural networks), and root-cause analysis (which can include probabilistic graphical models) can be trained on simulated data. The root-causes generated can be used for obtaining the corrective action. Corrective action can taken through simulation of the robotic path on the basis of the root causes found. The unit testing of the modules can be performed using nested cross-validation approaches on the data collected.

[0031] Figure 2 illustrates a process 200 for using controlled experiments to identify faults in a real-time analysis network in accordance with disclosed embodiments. In this figure, camera/ultrasound images 214 of the welds are sent to convolutional neural networks (CNN) 220. Sensor readings 216 of the welds, including weld parameters, are sent to autoencoder/recurrent neural networks (RNN) 222. The outputs of CNN 220 and autoencoder/RNN 222 are combined into fully connected layers 226. Fully connected layers 226 can then be analyzed using the deep neural networks to detect existence and type of defects 228.

[0032] Processes for robot weld simulation and weld defect data collection, in accordance with disclosed embodiments, can use finite element analysis methods to simulate the welding process to understand the physics and obtain data for the root cause analysis of defects. [0033] Disclosed embodiments model the complex interactions between electrical, thermal and mechanical phenomena as a multi-physics problem in three-dimensional space and solve them in a fully-coupled manner. The metallurgical effects are then considered after the finite element analysis process. The goal is to find some relationship between the heating and cooling rates and final structures of spot welds, based on some phase diagrams that are previously prepared.

[0034] Various embodiment can apply general mathematical equations that govern the electrical, thermal, and mechanical physics of the welds and welding processes. Electrical problems can be assumed to be governed by Quasi-Laplace equation with boundary condition v (c_e VV) = 0 in W,

c_e VV = — (V— g_±) on 3W, where c_e is a temperature dependent electric conductivity, V is the electrical potential, and g is the prescribed electrical potential on the boundary.

[0035] Thermal problems can be described by the heat transfer equation and boundary condition as pc_pT = V (kVT) + c_e V²(V) in W,

k VT = —k₀(T— g2) + h on 3W,

T = T₀ at t = 0, where p is the density, c_P is the specific heat per unit volume, T is the temperature and T its time derivative k is the heat conductivity, k₀ is the heat conductivity in the normal direction on the boundary, g₂ and h are the prescribed temperature and heat flux respectively on the boundary, and T₀ is the initial temperature.

For mechanical analysis, the governing equation is described in incremental form to relate the stress, strain and thermal expansion

As = ΰDe + CAT where \s and De are stress and strain increment respectively, matrix D is the elastic- plastic matrix, C is a thermal coefficient vector, and AT is the temperature differences.

[0036] The stress equilibrium equation is

V · <7 + pf - a where / is the body force, and a is the acceleration vector.

[0037] During the welding process of metal parts, usually multiple welds are needed to fully join the parts together with required strength. The robot arm follows a path that passes through all spots and welds one by one. Not only do the weld locations and tool path affect the welding quality at each weld, they also affect the overall structural behavior of the whole metal part. Therefore, in addition to simulating the multi-physics phenomena at a single weld, the system can also analyze an overall structural behavior of the whole metal part with multiple welds using a CWELD spot weld model.

[0038] The system can used equations given above iteratively or in a fully-coupled algorithm to provide temperature history predictions for the weld region. Typically, this thermal simulation is used directly to predict the diameter and penetration depth of the melt region. In order to accurately predict the propensity for defects, a more detailed simulation approach directly couples the electro-thermal-mechanical solution to the phase change of the metal (melting and solidification), the dynamics of the melt region, and the formation of metal microstructure during resolidification. Disclosed embodiments use the simulation tools to develop and validate detailed numerical simulations of weld formation. The geometries and microstructures predicted with this model provide a rich additional source of data for the development of weld defect detection algorithms in the associated project tasks.

[0039] The system can include a fully instrumented experimental testbed, built around a servo-controlled spot welding gun, and collect such data as welding current, voltage, resistance, electrode pressure, and temperature. The system can collect and analyze all relevant process responses, such as the simulation of process variations and disturbances for physical model validation and training of the learning algorithms. Metallurgical, mechanical and other joint properties will be assessed through customary methods.

[0040] As described above, the system can use deep-learning algorithms for weld defect detection. The challenges for weld defect detection include diverse data types, lack of accurate labels for historic data at customer site, and real-time defect prediction and customization at the customer-site through online learning.

[0041] Disclosed embodiments can use machine learning techniques such as fuzzy classification and clustering, support vector machines, k nearest neighbors, principal component analysis, and artificial neural networks. Not all defects may use the same set of features. Disclosed embodiments can implement a semi-supervised deep-learning approach which can leverage the unlabeled data.

[0042] Simulation data for factors such as robot path, part parameters, weld parameters and part movement can be used by the system in a semi-supervised learning process using autoencoders (context or variational) to predict future defects and defect type. The defects predicted can be processed by the root cause module to pre-emptively stop the defect from occurring. Then at real-time, the data from camera, ultrasound, temperature profile, and process parameters can be collected and sent to convolutional neural networks as illustrated in Figure 2 for predicting the run-time defects. The defects predicted at this stage is also sent to the root cause analysis tool to determine the root causes. The run-time defects are also fed back into the initial prediction as training data for improving the prediction network. Once the root causes are identified, are used for corrective action. The unlabeled data at the customer site can be used in a semi- supervised fashion for additional learning. The unlabeled data used for training can be fed into the network until a stable defect prediction is produced.

[0043] The system can use causal inference algorithms for root-cause analysis. Ignorance or misidentifi cation of the defect’s root causes can lead to a repetition and proliferation of defects. Current root-cause analysis techniques for welding-defects use statistical techniques which focus on single root causes. In an industrial welding setup, the defects may occur due to the interaction between multiple causes, as opposed to a single root cause. Counterfactual analysis inference models have been successfully used, for example, in the fields of genetics, medicine, and social science. The major benefits for these models includes their ability to distinguish factors that truly cause the defect from the factors that only correlate with it (the correlation might be, for example, due to common causes, also known as confounders).

[0044] In order to distinguish the causation from the correlation, disclosed embodiment can use causal graphical models and counterfactual reasoning. As opposed to purely statistical methods, which are based on probabilistic dependence between variables, causal inference methods introduce additional assumptions that allow them to look directly for causal relationships in data.

[0045] Using the data available through the simulations discussed above and causal graphical model learning algorithms, the system can create a causal graph that identifies causal relationships between the variables of interest. An instantiated causal graph enables the system to find the set of nodes (variables) that constitute the root causes of the defects. Using counterfactual reasoning, the system can predict“what would happen if’ possibilities by changing the values of the predicted root-cause variables.

[0046] Disclosed embodiments generate corrective paths for the spot welding robot, and can use a closed loop control algorithm for the robot welding. A welding path is usually generated based on the part geometry and location of the welding spots to avoid collision with any obstacles. The end effector of the robot welding arm is then supposed to move from one spot to another following the path. However, in practice, due to manufacturing tolerances the real part geometry may deviate from the model that is used to calculate the welding path. The system can generate a corrected path that compensates for such differences.

[0047] The system can use a close loop control algorithm to monitor the actual welding location and compared it with the predicted location on the pre-calculated tool path. The difference is then recorded and analyzed to calibrate the welding path, so that the error between the actual location and predicted location is minimized. This calibration process is done in real time, so as the welding proceeds and more data is collected, the more accurate the calibrated path becomes.

[0048] Disclosed embodiments improve on other systems in a number of ways, including improving weld time, cost per unit, and the quality of weld.

[0049] Figure 3 illustrates a flowchart of a process in accordance with disclosed embodiments that may be performed, for example, by a system as disclosed herein (referred to below as the“weld analysis system”).

[0050] The weld analysis system simulates a welding operation (302). The simulation can use one or more of a robotic path, part parameters, weld parameters, and part movements.

[0051] The weld analysis system analyzes a physical weld corresponding to the simulated welding operation (304). The analysis can include one or more of analyzing camera images, ultrasound images, sensor inputs from any of the various sensors described herein, and weld parameters.

[0052] The weld analysis system detects and predicts welding defects based on the simulated welding operation and the analysis of the physical weld (306). This can be performed by building one or more deep neural networks from the results of the simulated welding operation, such as features and other outputs, and the results of the physical weld analysis, such as the camera images, ultrasound images, sensor inputs from any of the various sensors described herein, weld parameters, or any other outputs.

[0053] The weld analysis system performs a root cause analysis, based on the detected and predicted welding defects, to identify at least one root cause of the welding defects (308). This can include building causal graphical models to evaluate the root cause, and can be performed using predicted defects based on the simulated welding operation and unpredicted, detected defects from the physical weld analysis. Further, the results of the root cause analysis can be fee back to the defect prediction and detection process, in particular to further train the deep neural network(s). [0054] The weld analysis system produces a corrective action based on the identified root cause(s) to remove the welding defects from subsequent physical welds (310). The corrective action can include, for example, corrected motion paths for robotic welders, defining corrected welding parameters, or others. Producing the corrective action can include performing path planning simulations to identify and test corrective actions.

[0055] The weld analysis system can thereafter perform physical welding operations according to the corrective action(s) (312).

[0056] Figure 4 illustrates a block diagram of a data processing system in which an embodiment can be implemented, for example as part of a weld analysis system particularly configured by software or otherwise to perform the processes as described herein, and in particular as each one of a plurality of interconnected and communicating systems as described herein. The data processing system depicted includes a processor 402 connected to a level two cache/bridge 404, which is connected in turn to a local system bus 406. Local system bus 406 may be, for example, a peripheral component interconnect (PCI) architecture bus. Also connected to local system bus in the depicted example are a main memory 408 and a graphics adapter 410. The graphics adapter 410 may be connected to display 411.

[0057] Other peripherals, such as local area network (LAN) / Wide Area Network / Wireless ( e.g . WiFi) adapter 412, may also be connected to local system bus 406. Expansion bus interface 414 connects local system bus 406 to input/output (I/O) bus 416. I/O bus 416 is connected to keyboard/mouse adapter 418, disk controller 420, and I/O adapter 422. Disk controller 420 can be connected to a storage 426, which can be any suitable machine usable or machine readable storage medium, including but not limited to nonvolatile, hard-coded type mediums such as read only memories (ROMs) or erasable, electrically programmable read only memories (EEPROMs), magnetic tape storage, and user-recordable type mediums such as floppy disks, hard disk drives and compact disk read only memories (CD-ROMs) or digital versatile disks (DVDs), and other known optical, electrical, or magnetic storage devices. [0058] Also connected to I/O bus 416 in the example shown is audio adapter 424, to which speakers (not shown) may be connected for playing sounds. Keyboard/mouse adapter 418 provides a connection for a pointing device (not shown), such as a mouse, trackball, trackpointer, touchscreen, etc. I/O adapter 422 can be connected to communicate with or control welding equipment 428, which can include welding robots, imagers, cameras, temperature sensors, ultrasound equipment, voltage sensors, current sensors, resistance sensors

[0059] Those of ordinary skill in the art will appreciate that the hardware depicted in Figure 4 may vary for particular implementations. For example, other peripheral devices, such as an optical disk drive and the like, also may be used in addition or in place of the hardware depicted. The depicted example is provided for the purpose of explanation only and is not meant to imply architectural limitations with respect to the present disclosure.

[0060] A data processing system in accordance with an embodiment of the present disclosure includes an operating system employing a graphical user interface. The operating system permits multiple display windows to be presented in the graphical user interface simultaneously, with each display window providing an interface to a different application or to a different instance of the same application. A cursor in the graphical user interface may be manipulated by a user through the pointing device. The position of the cursor may be changed and/or an event, such as clicking a mouse button, generated to actuate a desired response.

[0061] One of various commercial operating systems, such as a version of Microsoft Windows™, a product of Microsoft Corporation located in Redmond, Wash may be employed if suitably modified. The operating system is modified or created in accordance with the present disclosure as described.

[0062] LAN/ WAN/Wireless adapter 412 can be connected to a network 430 (not a part of data processing system 400), which can be any public or private data processing system network or combination of networks, as known to those of skill in the art, including the Internet. Data processing system 400 can communicate over network 430 with server system 440, which is also not part of data processing system 400, but can be implemented, for example, as a separate data processing system 400.

[0063] Of course, those of skill in the art will recognize that, unless specifically indicated or required by the sequence of operations, certain steps in the processes described above may be omitted, performed concurrently or sequentially, or performed in a different order.

[0064] Those skilled in the art will recognize that, for simplicity and clarity, the full structure and operation of all data processing systems suitable for use with the present disclosure is not being depicted or described herein. Instead, only so much of a data processing system as is unique to the present disclosure or necessary for an understanding of the present disclosure is depicted and described. The remainder of the construction and operation of data processing system 100 may conform to any of the various current implementations and practices known in the art.

[0065] It is important to note that while the disclosure includes a description in the context of a fully functional system, those skilled in the art will appreciate that at least portions of the mechanism of the present disclosure are capable of being distributed in the form of instructions contained within a machine-usable, computer-usable, or computer- readable medium in any of a variety of forms, and that the present disclosure applies equally regardless of the particular type of instruction or signal bearing medium or storage medium utilized to actually carry out the distribution. Examples of machine usable/readable or computer usable/readable mediums include: nonvolatile, hard-coded type mediums such as read only memories (ROMs) or erasable, electrically programmable read only memories (EEPROMs), and user-recordable type mediums such as floppy disks, hard disk drives and compact disk read only memories (CD-ROMs) or digital versatile disks (DVDs).

[0066] Although an exemplary embodiment of the present disclosure has been described in detail, those skilled in the art will understand that various changes, substitutions, variations, and improvements disclosed herein may be made without departing from the spirit and scope of the disclosure in its broadest form. [0067] None of the description in the present application should be read as implying that any particular element, step, or function is an essential element which must be included in the claim scope: the scope of patented subject matter is defined only by the allowed claims. Moreover, none of these claims are intended to invoke 35 USC §112(f) unless the exact words "means for" are followed by a participle. The use of terms such as (but not limited to) “mechanism,” “module,” “device,” “unit,” “component,” “element,” “member,” “apparatus,” “machine,” “system,” “processor,” or“controller,” within a claim is understood and intended to refer to structures known to those skilled in the relevant art, as further modified or enhanced by the features of the claims themselves, and is not intended to invoke 35 U.S.C. §112(f).

Claims

WHAT IS CLAIMED IS:

1. A method (300) performed by a weld analysis system (400), comprising:

simulating (302) a welding operation (102) by the weld analysis system (400); analyzing (304) a physical weld (112) corresponding to the simulated welding operation, by the weld analysis system (400);

detecting and predicting (306) welding defects, by the weld analysis system (400), based on the simulated welding operation and the analysis of the physical weld;

performing a root cause analysis (124, 308), by the weld analysis system (400) and based on the detected and predicted welding defects, to identify at least one root cause of the welding defects; and

producing a corrective action (128, 310), by the weld analysis system (400), based on the identified at least one root cause to remove the welding defects from subsequent physical welds.

2. The method of claim 1, further comprising performing physical welding operations (312) according to the corrective action.

3. The method of any of claims 1-2, wherein the simulated welding operation (102) uses one or more of a robotic path, part parameters, weld parameters, and part movements.

4. The method of any of claims 1-3, wherein the analysis of the physical weld includes one or more of analyzing camera images (114), analyzing ultrasound images (114), analyzing sensor inputs (116), and analyzing weld parameters (116).

5. The method of any of claims 1-4, wherein detecting and predicting welding defects is performed by building one or more deep neural networks (120, 122) from results of the simulated welding operation and results of the physical weld analysis.

6. The method of any of claims 1-5, wherein the root cause analysis includes building causal graphical models (126) using predicted defects based on the simulated welding operation and unpredicted, detected defects from the physical weld analysis.

7. The method of any of claims 1-6, wherein the corrective action includes corrected motion paths for robotic welders.

8. A weld analysis system (400) comprising:

a processor (402); and

an accessible memory (408), the weld analysis system particularly configured to perform a process as in any of claims 1-7.

9. A non-transitory computer-readable medium (408, 426) storing executable instructions that, when executed, cause one or more weld analysis systems (400) to perform a process as in any of claims 1-7.