Industrial robot edge rolling process quality detection method and system based on supervised learning

Abstract

The invention discloses an industrial robot edge rolling process quality detection method and system based on supervised learning, and belongs to the technical field of intelligent manufacturing and industrial big data analysis. The method comprises the following steps: acquiring multi-dimensional time sequence track data in the operation process of the binding robot; dividing the time sequence data into independent track segments; for each track segment, extracting four core statistical features of a mean value, a standard deviation, a maximum value and a minimum value, and further constructing a feature vector; training a classification model by using the feature vectors of which the states are marked in the historical data, and establishing a nonlinear mapping model from the features to the states; and finally, inputting the new trajectory feature vector into the model, and outputting a classification result and a prediction probability of the process state in real time. According to the method, the historical label data is fully utilized, high-precision and high-efficiency automatic online monitoring and classification of the known fault mode binding process state are realized through lightweight feature engineering and an efficient classification algorithm, and an effective technical means is provided for accurate identification and predictive maintenance of the known fault mode.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of intelligent manufacturing and industrial big data analysis, and particularly relates to a method and system for classifying and identifying the state of robot trajectory data of a rolling edge process in an industrial production process based on historical label data and a machine learning classification algorithm. BACKGROUND

[0002] In modern industrial manufacturing processes, the rolling edge process is a key link that determines the appearance quality of a vehicle body. This process is usually performed by an industrial robot, which precisely controls the end roller to press the edge of a vehicle body sheet metal part along a predetermined trajectory, and the quality directly affects the appearance consistency, sealing performance, and structural strength of the vehicle body. Currently, the monitoring of the rolling edge quality relies mainly on post-optical detection or manual sampling inspection. This approach has the following drawbacks:

[0003] (1) Lagging: It is impossible to perform real-time monitoring and early warning during the rolling edge process, and when quality problems are found, a large number of defective products have already been produced.

[0004] (2) High cost: Precise optical detection equipment requires a huge investment, and manual detection is inefficient and subject to subjective factors.

[0005] (3) Low data utilization: The rolling edge robot generates a large amount of process data during operation, which contains valuable information reflecting the process state, but there is currently a lack of a technical means to automatically learn from this data and convert it into a classification model that can be applied online in real time. Therefore, developing a supervised learning solution that can fully exploit and utilize the value of existing labeled data is of great significance to improving the intelligent level of quality control.

[0006] Therefore, there is an urgent need in the industry for a method that can utilize real-time data during the rolling edge process to quickly, accurately, and cost-effectively determine the state of the process trajectory, thereby promptly identifying abnormalities and preventing defects. SUMMARY

[0007] The present application aims to solve the problems of lagging quality monitoring, high cost, and insufficient utilization of process data in the rolling edge process in the prior art, and provides a technical solution that can achieve real-time, high-precision, and automated online monitoring and judgment of the state of the rolling edge process.

[0008] To achieve the above-mentioned purpose, the present application provides an industrial robot rolling edge process quality detection method based on supervised learning. The core of this method is to transform the high-dimensional, unstructured time series trajectory data generated during robot operation into structured feature vectors through feature engineering, and to establish a mapping relationship between the feature vectors and the process state using a machine learning model.

[0009] A method for detecting the quality of industrial robot rolling process based on supervised learning, comprising:

[0010] S1: Data acquisition, obtaining multi-dimensional time sequence trajectory data of one or more working cycles of the robot end effector in the rolling process;

[0011] S2: Data segmentation, automatically segmenting the multi-dimensional time sequence trajectory data into trajectory segments corresponding to single, independent rolling tasks through a periodic detection algorithm;

[0012] S3: Feature extraction, for each trajectory segment, a set of statistical features is calculated, and the set of statistical features is combined into a feature vector, wherein the statistical features at least include quantities describing the trend, dispersion and distribution form of the data set;

[0013] S4: Model training and prediction, providing a training data set containing a plurality of feature vectors and their corresponding known state labels, training a classification model using the training data set to obtain a state discrimination model; when predicting, a newly acquired trajectory segment generates a feature vector through the steps of constructing a feature vector according to steps S1-S3, and inputs the feature vector into the state discrimination model, and outputs the process state classification result and the prediction probability corresponding to the trajectory.

[0014] Further, the multi-dimensional time sequence trajectory data in the S1 step is derived from at least two robots in different physical positions to construct a state discrimination model that can be applied to multiple robots.

[0015] Further, the robot working cycle obtained in the S1 step is defined by a periodic signal marker in the robot control program, and one cycle corresponds to the completion of a single rolling task for a workpiece. The multi-dimensional time sequence trajectory data obtained in each cycle includes a sequence of three-dimensional coordinates (X, Y, Z) of the robot end flange center, a sequence of angles of the six joints, and a sequence of torques of the six joints.

[0016] Further, in the S3 step, the statistical features include calculating the mean, standard deviation, maximum value and minimum value for each original variable to form a feature vector.

[0017] Further, in the S4 step, during the training of the classification model, a plurality of different types of machine learning classification models are trained in parallel; the performance indicators of each machine learning classification model are evaluated in a cross-validation manner, and according to a predetermined selection criterion, the machine learning classification model with the best performance is selected from the candidate machine learning classification models as the final state discrimination model.

[0018] Further, the optimization strategy of the model training module in the S4 step is the Bayesian optimization algorithm.

[0019] Further, the performance indicators in the S4 step include one or more of accuracy, precision, recall, F1 score, or AUC value.

[0020] A supervised learning-based industrial robot rolling process quality detection system, comprising a processor and a memory, the memory storing computer executable instructions, the processor executing the instructions to configure the system to:

[0021] A data acquisition module for collecting multi-dimensional time series trajectory data of a robot end effector in a rolling process;

[0022] A data processing module for segmenting the multi-dimensional time series trajectory data into trajectory segments corresponding to single, independent rolling tasks;

[0023] A feature vector construction module for calculating a set of statistical features for each trajectory segment and combining the set of statistical features into a feature vector;

[0024] A model training and prediction module for training a classification model based on historical labeled data and generating a state discrimination model, and receiving a feature vector of a new trajectory segment from the feature vector construction module, and outputting a classification result of the process state corresponding to the trajectory segment;

[0025] An alarm and linkage control module for receiving the classification result directly output by the optimal model, and performing at least one of the following operations when the classification result is abnormal: triggering an alarm on a human-machine interaction interface, automatically marking the current workpiece as to be rechecked, or sending a control instruction to a production line central control system.

[0026] A computer-readable storage medium having computer instructions for causing a processor to execute the above-mentioned industrial robot rolling process quality detection method.

[0027] The core idea of the present application is summarized as follows:

[0028] Data acquisition step: acquiring multi-dimensional time series trajectory data of one or more work cycles of a robot end effector in a rolling process.

[0029] Data processing step: segmenting the time series trajectory data into trajectory segments corresponding to independent rolling tasks.

[0030] Feature vector construction step: calculating a set of statistical features for each trajectory segment and combining the set of statistical features into a feature vector.

[0031] Model training and prediction steps: Provide a training dataset containing multiple feature vectors and their corresponding known state labels, train a classification model using the training dataset, and input new feature vectors into the state discrimination model to output a classification result representing the process state corresponding to the trajectory segment.

[0032] Accordingly, the present invention also provides a system for implementing the above method, the system comprising: a data acquisition module, a data processing module, a feature vector construction module, a model training and prediction module, and an alarm and linkage module. These modules work together to execute the various steps of the above method.

[0033] Furthermore, the present invention also provides a computer-readable storage medium on which a computer program stored is executed by a processor to implement the above-described method.

[0034] Compared with the prior art, the present invention has the following significant advantages:

[0035] (1) Real-time online monitoring: By analyzing process data in real time, quality judgment can be moved from post-inspection to in-process monitoring, which can detect process drift or sudden abnormalities in time, effectively reduce the generation of defective products, and realize predictive maintenance.

[0036] (2) Reduced detection costs and complexity: This invention does not require additional expensive optical detection hardware, but makes full use of the data generated by the robot controller itself, which is a low-cost, non-invasive, and high-efficiency quality control solution.

[0037] (3) Improve the accuracy and robustness of judgment: By transforming dynamic time series data into stable statistical feature vectors, the interference of data noise is effectively reduced; combined with the excellent classification performance and generalization ability of machine learning algorithms, the high accuracy and stability of state judgment are ensured.

[0038] (4) Provide data-driven decision support: The model can not only judge the status, but its output can also provide accurate data for process optimization and fault diagnosis, and promote the intelligentization of the production process. Attached Figure Description

[0039] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0040] Figure 1 This is a flowchart of a method for quality inspection of industrial robot hemming process based on supervised learning, according to Embodiment 1 of the present invention.

[0041] Figure 2 This is for comparing the F1 scores of different models provided in Embodiment 1 of the present invention.

[0042] Figure 3 The OOB error curve of the random forest model provided in Embodiment 1 of the present invention.

[0043] Figure 4 This is a schematic diagram of a system for determining the state of rolled edge data trajectory based on a supervised learning model according to Embodiment 2 of the present invention. Detailed Implementation

[0044] To enable those skilled in the art to better understand the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are merely some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort should fall within the scope of protection of the present invention.

[0045] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this application. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0046] It should be noted that the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of methods and systems according to various embodiments of this disclosure. It should be noted that each block in a flowchart or block diagram may represent a module, segment, or portion of code, which may include one or more executable instructions for implementing the logical functions specified in the various embodiments. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than that shown in the drawings. For example, two consecutively represented blocks may actually be executed substantially in parallel, or they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the flowcharts and / or block diagrams, and combinations of blocks in the flowcharts and / or block diagrams, may be implemented using a dedicated hardware-based system that performs the specified functions or operations, or using a combination of dedicated hardware and computer instructions.

[0047] Example 1: A Quality Inspection Method for Hemming Process of Industrial Robots Based on Supervised Learning

[0048] like Figure 1 As shown, this embodiment provides a method for quality inspection of industrial robot hemming process based on supervised learning, including the following steps:

[0049] Step S101: Data acquisition. Obtain multi-dimensional time-series trajectory data of one or more work cycles of the robot end effector in the hemming process. The data can be acquired in real time at a preset frequency or read in batches from a historical database.

[0050] The multidimensional time-series trajectory data includes at least:

[0051] The sequence of three-dimensional coordinates (X, Y, Z) of the robot end flange center in the world coordinate system or tool coordinate system.

[0052] Angle sequence of the robot's six joints (J1-J6).

[0053] Torque sequence of the robot's six joints (mtq1-mtq6).

[0054] To improve the model's generalization ability and robustness, a key implementation method is to collect data from robots at at least two different physical locations on the production line and use this data together for subsequent model training. The model trained in this way will not overfit to minor characteristics of a specific hardware component and can be more broadly applied to multiple workstations.

[0055] Step S102: Data segmentation. The raw data is a continuous time-series stream. An automated periodic detection unit identifies specific signals in the robot control program, segmenting the raw data into "trajectory segments" corresponding to a single, independent rolling task. For example, the process from the start of rolling to the completion of rolling the edge of a car door cover constitutes a trajectory segment. This unit can automatically perform the segmentation using task start / end flags or by analyzing the periodic characteristics of the trajectory data, without manual intervention.

[0056] Step S103: Feature extraction, which is one of the core technical points of this invention. In a preferred embodiment of this invention, for the trajectory segment, four statistical features—mean, standard deviation, maximum value, and minimum value—are calculated for each original variable to form a feature vector with robust feature dimensions.

[0057] This design takes into account that in production environments, the number of clearly labeled fault state samples is usually extremely limited. Using too many dimensions of features can easily lead to the curse of dimensionality, causing the model to memorize noise from the training set instead of general patterns, thus severely impairing its generalization performance on unknown data. This invention minimizes the feature dimensionality by carefully selecting four core features with clear physical meaning and high discriminative power, fundamentally ensuring the robustness of the model under scarce samples.

[0058] Secondly, the mean reflects the average level of the trajectory, the standard deviation characterizes its fluctuation stability, and the maximum and minimum values ​​can capture abnormal peak events. These four features together constitute the most comprehensive and robust feature set describing the trajectory state, while requiring very low computational resources, thus meeting the production line's requirements for high efficiency and low cost in real-time monitoring systems.

[0059] Step S104: Model Training. The core of this step is to utilize the statistical features of labeled normal / fault trajectory samples to train a machine learning classification model that is most suitable and robust in the current data environment as the state discrimination model, avoiding the performance risks associated with subjective model selection. The detailed process is as follows:

[0060] 1. Select K-Nearest Neighbors (KNN), Support Vector Machine (SVM), Neural Network (NN), and Random Forest (RF) models as candidate model sets to cover different model hypothesis spaces.

[0061] 2. Constructing the training dataset: For a newly acquired edge-rolling trajectory from the production line, the system first executes processes S101-S103 in real time to generate its feature vector, and assigns supervisory labels to each feature vector with reference to the original file. All cyclic features are concatenated into a sample table, with the State label column representing a categorical variable.

[0062] 3. Model Training and Hyperparameter Optimization: The same training steps were performed for each model. To obtain the best model performance, Bayesian optimization was used for automatic optimization during training, and five-fold cross-validation was used to prevent overfitting. The performance metrics of all models were compared. Considering that in industrial scenarios, the number of abnormal samples is far less than the number of normal samples, potentially leading to data imbalance, this embodiment uses the F1 score as the core selection criterion. This is because it simultaneously considers the model's precision and recall, and more accurately reflects the model's performance on imbalanced datasets than precision alone. Finally, the random forest model with the highest F1 score was selected as the final model for deployment.

[0063] This embodiment directly utilizes labeled samples for training, aiming to minimize classification errors. The classification boundary is automatically learned through data-driven methods, and generalization ability is ensured through Bayesian optimization and cross-validation. The model output consists of classification predictions and posterior probabilities. The Out-of-Bag (OOB) error curve of the random forest model provides an interpretable representation of the classification criteria, facilitating process regulation and root cause analysis. Figure 3 As shown.

[0064] Example 2: A Quality Inspection System for Hemming Process of Industrial Robots Based on Supervised Learning

[0065] This embodiment provides a supervised learning-based industrial robot hemming process quality inspection system. This system can be seamlessly integrated into existing automotive manufacturing production lines. Its physical and software architecture is as follows: Figure 4 As shown, it mainly consists of three parts: a data acquisition terminal, a model service and processing center, and an application alarm terminal.

[0066] 1. System Hardware Architecture and Deployment

[0067] Data Acquisition Terminal: An industrial edge computing gateway is deployed next to each hemming robot control cabinet. This gateway communicates directly with the robot controller (PLC) via the industrial Ethernet protocol. It is configured to read and cache the three-dimensional coordinates (X, Y, Z) sequence of the robot's end flange center and the angle and torque sequences of the six joints at a high speed of 20ms, totaling 15 dimensions of timing data. Simultaneously, the gateway also listens to the signal bits in the PLC that mark "hemming task start" and "hemming task end".

[0068] Model Service and Processing Center: A high-performance industrial server is deployed in the factory's server room. This server serves as the system's "brain," housing data receiving services, a real-time computing engine, a model training platform, and a database. It connects to all front-end edge computing gateways via the factory's local area network.

[0069] Application alarm terminal: Deploy the system's web-based visual front-end interface on the workstation monitoring screen (HMI) next to the production line or on the monitoring screen in the central control room.

[0070] 2. System Workflow

[0071] The workflow of this system is divided into two stages: online real-time judgment and offline model training and updating.

[0072] (a) Online real-time judgment process:

[0073] Step 1: Data Acquisition and Transmission. When the production line starts and the robot begins its hemming operation on a car door, the edge computing gateway detects the task start signal and immediately begins recording 15-dimensional time-series trajectory data. When the task end signal appears, the gateway sends the data packet of this complete trajectory segment to the model service and processing center via the MQTT protocol.

[0074] Step Two: Feature Extraction and Feature Vector Construction. Upon receiving the data packet, the data processing module immediately starts. It quickly calculates the mean, standard deviation, maximum, and minimum values ​​for each of the 12 variables. A high-dimensional time-series data segment is thus transformed into a 15 * 4 = 60-dimensional feature vector.

[0075] Step 3: Model Prediction and Result Output. The 60-dimensional feature vector is fed into the deployed random forest state discrimination model, which was selected through optimization in Example 1. The model completes the calculation within 10ms and outputs two results: one is the state classification (e.g., "normal", "Class A anomaly - trajectory deviation", "Class B anomaly - torque mutation"), and the other is the confidence probability of the judgment (e.g., 98.5%).

[0076] Step 4: Alarm and Linkage Control. The alarm and linkage control module receives the prediction result. If the result is "normal," the system remains silent, only logging in the database; if the result is any "abnormal" category, the system will immediately execute the linkage operation:

[0077] 1. Push information to the application alarm terminal and pop up a red alarm window on the HMI screen with the content "Robot No. 3 rolling trajectory abnormal, type: torque mutation, confidence level 98.5%".

[0078] 2. Send a command to the production line central control system (PLC) via the network to mark the currently completed door workpiece as "pending re-inspection" in the system so that it can be subject to key manual inspection in subsequent quality inspection stages.

[0079] 3. Store complete event records, including raw data, feature vectors, and model results, in the database for subsequent process analysis.

[0080] (b) Offline model training and update process:

[0081] Quality inspectors will regularly inspect the workpieces marked as "pending re-inspection" and mark the final confirmed fault type (or confirmed no fault) in the system.

[0082] Every month, or once a sufficient number of new labeled samples have been accumulated, engineers will start the model training platform on the server. The platform will automatically pull all the labeled trajectory data from this period and repeat the steps in Example 1, namely, using cross-validation and Bayesian optimization to retrain and evaluate multiple candidate models, including random forests, support vector machines, and neural networks.

[0083] If training discovers a new model or a new set of hyperparameters whose F1 score is significantly higher than the currently deployed online model, the system will prompt engineers to perform a one-click update, deploying the better state discrimination model into the online real-time judgment process, thereby achieving continuous iteration and optimization of the model.

[0084] Through the above design, this embodiment constructs a closed-loop intelligent monitoring system that integrates data acquisition, real-time analysis, intelligent alarm, and model self-optimization, thus fully implementing the method of the present invention in actual industrial scenarios.

Claims

1. A method for detecting the quality of a rolling process of an industrial robot based on supervised learning, characterized in that, The method comprises the following steps: S1: data acquisition, obtaining multi-dimensional time sequence trajectory data of one or more working cycles of a robot end effector in a rolling process; S2: data segmentation, automatically segmenting the multi-dimensional time sequence trajectory data into trajectory segments corresponding to single, independent rolling tasks through a periodic detection algorithm; S3: feature extraction, for each trajectory segment, calculating a set of statistical features, and combining the set of statistical features into a feature vector, wherein the statistical features at least include quantities describing the trend, dispersion degree and distribution form of the data set; S4: model training and prediction, providing a training data set containing a plurality of feature vectors and their corresponding known state labels, training a classification model using the training data set to obtain a state discrimination model; in prediction, a newly obtained trajectory segment generates a feature vector through the steps of constructing a feature vector in steps S1-S3, and the feature vector is input into the state discrimination model to output the process state classification result and the prediction probability corresponding to the trajectory.

2. The method of claim 1, wherein, The multi-dimensional time sequence trajectory data in the S1 step is derived from at least two robots in different physical positions to construct a state discrimination model that can be applied to multiple robots.

3. The method of claim 1, wherein, The robot working cycle obtained in the S1 step is defined by a periodic signal marker in the robot control program, and one cycle corresponds to the completion of a single rolling task for a workpiece. The multi-dimensional time sequence trajectory data obtained in each cycle includes a sequence of three-dimensional coordinates (X, Y, Z) of the center of the robot end flange, a sequence of angles of the six joints, and a sequence of torques of the six joints.

4. The method of claim 1, wherein, In the S3 step, the statistical features include calculating the mean, standard deviation, maximum value and minimum value of each original variable to form a feature vector.

5. The method of claim 1, wherein, In the S4 step, a plurality of different types of machine learning classification models are trained in parallel during the training of the classification model. The performance indicators of each machine learning classification model are evaluated in a cross-validation manner, and the best machine learning classification model is selected as the final state discrimination model according to a preset optimization criterion.

6. The method of claim 1, wherein, The optimization strategy of the model training module in the S4 step is the Bayesian optimization algorithm.

7. The method of claim 1, wherein, The performance indicators in the S4 step include one or more of accuracy, precision, recall, F1 score or AUC value.

8. A supervised learning based industrial robot overedge process quality detection system, characterized in that, The system comprises a processor and a memory, and the memory stores computer executable instructions. When the processor executes the instructions, the system is configured to: a data acquisition module for acquiring multi-dimensional time sequence trajectory data of a robot end effector in a rolling process; a data processing module for segmenting the multi-dimensional time sequence trajectory data into trajectory segments corresponding to single, independent rolling tasks; a feature vector construction module for calculating a set of statistical features for each trajectory segment and combining the set of statistical features into a feature vector; The model training and prediction module is configured to train a classification model based on historical labeled data and generate a state discrimination model, and receive the feature vector of a new trajectory segment from the feature vector construction module and output a classification result of a process state corresponding to the trajectory segment. The alarm and linkage control module is configured to receive the classification result directly output by the optimal model, and when the classification result is abnormal, perform at least one of the following operations: trigger an alarm on a human-machine interaction interface, automatically mark the current workpiece as to be rechecked, or send a control instruction to a central control system of a production line.

9. A computer-readable storage medium, characterized in that, The computer readable storage medium has computer instructions for causing the processor to implement the industrial robot edge rolling process quality detection method of any one of claims 1-7 when executed.

Images