State detection and fault diagnosis method for mechanical arm

Abstract

The invention discloses a state detection and fault diagnosis method for a mechanical arm. The method comprises the steps that deformation data and temperature data of all joints of the mechanical arm are obtained through a film sensor; preprocessing the deformation data and the temperature data to obtain input data; constructing a prediction model of a hybrid neural network based on a long-short-term memory network and in combination with a convolutional neural network; inputting the input data into the prediction model, wherein the prediction model outputs a corresponding prediction deformation quantity based on forward propagation; and the state of the mechanical arm is judged through the predicted deformation quantity by adopting a loss function. By means of the method, compared with the mode that the state is judged directly through detection parameters, the method is more accurate and has predictability, meanwhile, the mechanical arm under the complex condition can be better and accurately judged in combination with an algorithm, and misjudgment is effectively avoided.

Description

Technical Field

[0001] The present application relates to the technical field of robotic arms, and in particular to a state detection and fault diagnosis method for robotic arms. Background Art

[0002] In today's industrial production process, robotic arms are widely used in automobile manufacturing, electronics and electrical, metal processing, food and drug production, logistics warehousing, precision processing and other fields due to their high efficiency, high precision, multiple input and output characteristics. In addition, due to their advantages in repetitive labor, they can save a lot of manpower, improve production efficiency, realize the automation of industrial production and greatly improve production efficiency.

[0003] As technology continues to innovate, robotic arms integrate more and more functions, making their systems more complex. The resulting uncertainty from parameter perturbations and external interference is also increasing, driving up their costs, maintenance, and upgrades. Consequently, there is a growing demand for testing and diagnostics of robotic arms to ensure their stability in industrial production applications, indirectly improving the stability and reliability of industrial production.

[0004] Existing robotic arm detection and diagnosis often involves direct comparison and analysis of collected parameters. This method is relatively crude and can only make judgments on situations where obvious abnormalities have occurred or misjudge heavy-load operations. It cannot cope with complex working situations or predict potential changes. The judgment logic is simple and the accuracy is low. Summary of the Invention

[0005] This application mainly provides a state detection and fault diagnosis method for a robotic arm to improve the accuracy of judging the condition of the robotic arm and improve the prediction ability.

[0006] To solve the above technical problems, a technical solution adopted in this application is to provide a state detection and fault diagnosis method for a robotic arm, comprising the steps of: S10: Obtain deformation data and temperature data of each joint of the robotic arm through thin film sensors; S20: Preprocessing the deformation data and the temperature data to obtain input data; S30: Combine the long short-term memory network and the convolutional neural network, initialize the model parameters and build a prediction model of the hybrid neural network based on the input data S40: The prediction model outputs the corresponding predicted shape variable using forward propagation according to the real-time data; S50: Using a loss function to determine the real-time state of the robotic arm through the predicted deformation amount.

[0007] In a possible implementation, the step of preprocessing the deformation data and the temperature data to obtain input data includes: S21: Cleaning the collected deformation data and temperature data and checking for missing values. If the missing values ​​do not exceed a threshold, data filling is performed; if the missing values ​​exceed a threshold, the corresponding data is deleted. S22: Output the cleaned data.

[0008] In a possible implementation, the step of preprocessing the deformation data and the temperature data to obtain input data includes: S23: Perform data enhancement on the deformation data and the temperature data to obtain enhanced data.

[0009] In a possible implementation, the step of preprocessing the deformation data and the temperature data to obtain input data includes: S24: Divide the cleaned data and the enhanced data into a training set, a validation set, and a test set according to a predetermined ratio.

[0010] In one possible implementation, the steps of combining a long short-term memory network with a convolutional neural network, initializing the model parameters, and constructing a prediction model of the hybrid neural network based on the input data include: S31: The hybrid neural network includes a convolutional neural network for extracting features from the input data and a long short-term memory network for performing time series analysis on the features.

[0011] In one possible implementation, the prediction model outputs the corresponding predicted shape variable using forward propagation based on real-time data, including: S41: Adjusting the prediction model based on the evaluation result of the loss function, where the loss function is defined by the mean square error.

[0012] In one possible implementation, the prediction model outputs the corresponding predicted shape variable using forward propagation based on real-time data, including: S42: Obtain prediction errors based on the evaluation results, use a back propagation algorithm to transmit the prediction errors back to neurons in each layer, and update the prediction model according to the gradient descent principle.

[0013] In a possible implementation, the step of using a loss function to determine the real-time state of the robotic arm through the predicted deformation includes: S51: obtaining an actual deformation value based on the deformation data, and quantifying, comparing, and evaluating the difference between the predicted deformation value and the actual deformation value using the loss function.

[0014] In a possible implementation, the step of using a loss function to determine the real-time state of the robotic arm through the predicted deformation includes: S52: combining the evaluation of the loss function with the data characteristics under different working conditions to establish data change patterns under different working conditions; S53: The prediction model dynamically selects a corresponding determination result based on the data change rule.

[0015] In a possible implementation, the step of using a loss function to determine the real-time state of the robotic arm through the predicted deformation includes: S54: When the state of the robotic arm is determined to be a fault, the determination result, fault type, fault location, fault severity, and maintenance suggestions are fed back to the operator through a visual interface.

[0016] The beneficial effects of the present application are: different from the existing technology, the present application discloses a state detection and fault diagnosis method for a robotic arm. After obtaining various parameters of the robotic arm through a thin film sensor, the deformation of the robotic arm is predicted and its state is judged through the prediction model of the long short-term memory network and the convolutional neural network. This method is more accurate and predictive than directly judging the state by detecting parameters. At the same time, combined with the algorithm, it can better make accurate judgments on the robotic arm in complex situations and effectively avoid misjudgment. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present application. Those skilled in the art can also derive other drawings based on these drawings without inventive work, among which: Figure 1 It is a flow chart of a method for state detection and fault diagnosis of a robotic arm in one embodiment of the present application. DETAILED DESCRIPTION

[0018] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0019] The terms "first", "second" and "third" in the embodiments of the present application are only used for descriptive purposes and are not to be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, a feature defined as "first", "second" and "third" may explicitly or implicitly include at least one of the features. In the description of the present application, the meaning of "plurality" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device comprising a series of steps or units is not limited to the listed steps or units, but may optionally also include steps or units that are not listed, or may optionally also include other steps or units inherent to these processes, methods, products or devices.

[0020] References to "embodiments" herein mean that a particular feature, structure, or characteristic described in connection with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various places in the specification does not necessarily refer to the same embodiment, nor does it constitute an independent or alternative embodiment that is mutually exclusive of other embodiments. It is understood, both explicitly and implicitly, by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0021] See also Figure 1 , the embodiment of the present application proposes a state detection and fault diagnosis method for a robotic arm, comprising the steps of: S10: Obtain deformation data and temperature data of each joint of the robotic arm through thin film sensors; S20: Preprocess the deformation data and temperature data to obtain input data; S30: Combine the long short-term memory network and the convolutional neural network, initialize the model parameters and build a prediction model of the hybrid neural network based on the input data; S40: The prediction model uses forward propagation to output the corresponding predicted shape variable based on the real-time data; S50: Using the loss function to predict the deformation variable to determine the real-time state of the robot arm.

[0022] Specifically, in step S10, thin-film sensors are installed at key locations on the robotic arm, including its joints. These sensors can collect physical parameters such as surface stress, strain, and temperature, and feature high precision, excellent creep resistance, and strong anti-interference capabilities. Furthermore, laser sensors, capacitive sensors, and other sensors can be added to detect the displacement of the robotic arm. Combined with the detection results of multiple sensors, the collected deformation or temperature data can be more accurately acquired and interpreted.

[0023] In step S20, the deformation and temperature data are preprocessed to meet the input requirements of the prediction model. Preprocessing includes denoising and normalization. Denoising removes or corrects noise and anomalies in the data to ensure that overall data changes are reasonable. This prevents interference caused by detection issues, errors, hardware defects, and other factors, which could indirectly lead to misjudgment of the robot arm. Normalization ensures that the data format is uniform and usable for subsequent processing.

[0024] In step S30, a hybrid neural network prediction model is constructed by combining a long short-term memory (LSTM) network with a convolutional neural network (CNN). The model outputs the deformation of the robotic arm based on the acquired deformation and temperature data. The model is constructed and trained based on existing data. In this embodiment, the existing input data is divided into a training set, a validation set, and a test set in a ratio of 7:2:1. These are used for model parameter training, model parameter adjustment, and model performance evaluation, respectively.

[0025] Initialize and set the model parameters, including the number of LSTM network layers, the number of CNN network layers, the number of neurons, the learning rate, the number of iterations, etc., and then adjust and optimize the parameters based on the convergence of the model after iteration. In this embodiment, the LSTM network includes two layers, each including 64 neurons, and the CNN network includes a convolutional layer and a pooling layer. The training set allows the prediction model to learn the patterns in the data, and the validation set is used to adjust some parameters in the model, such as the learning rate, the number of neurons, etc., while monitoring whether the prediction model is overfitting. Finally, the test set is used to evaluate the model performance, especially the prediction accuracy of new data that has not been processed.

[0026] In step S40, the model obtained after training processes the input data obtained by preprocessing the newly collected real-time data, and then outputs the predicted deformation variable of the robotic arm after passing through each layer in sequence through forward propagation.

[0027] In step S50, a loss function is introduced to measure the difference between the predicted deformation and the actual deformation, thereby achieving early fault warning and reducing the dynamic impact of multiple factors. Furthermore, a comparison of the actual deformation with a threshold can be combined to achieve dual screening of the robot arm's status, thereby improving the accuracy of the judgment. The actual deformation is calculated based on deformation data and temperature data.

[0028] Compared with the traditional method of state judgment based on a single deformation variable itself, this embodiment can perform dynamic analysis based on multi-source data, effectively targeting hidden faults. At the same time, the prediction model can adaptively predict various working conditions, reduce noise disturbances, and effectively improve the judgment accuracy.

[0029] In one embodiment, the step of preprocessing the deformation data and the temperature data to obtain input data includes: S21: Clean the collected deformation data and temperature data and check for missing values. If the missing values ​​do not exceed the threshold, fill in the data; if the missing values ​​exceed the threshold, remove the corresponding data. S22: Output the cleaned data.

[0030] Specifically, the preprocessing of deformation data and temperature data includes data cleaning, checking whether there are missing values ​​in each segment of data, and setting a threshold for the missing values. When the missing values ​​are small and below the threshold, the weighted mean of adjacent data or interpolation method is used to fill the data; when the missing values ​​are large and exceed the threshold, the entire segment of data is directly deleted.

[0031] Furthermore, the data is identified to remove outliers. The interquartile range of the data is calculated, and data outside the normal range is considered as outliers and corrected or deleted to ensure the quality of the input data. After the data is cleaned, it is output.

[0032] In one embodiment, the step of preprocessing the deformation data and the temperature data to obtain input data includes: S23: Perform data enhancement on the deformation data and the temperature data to obtain enhanced data.

[0033] Specifically, the preprocessing of deformation data and temperature data also includes data enhancement, which involves performing operations such as time series translation and scaling on the collected deformation data to generate new samples for simulating strain changes at different times; or performing low-proportion scaling on the collected temperature data to generate samples with different degrees of temperature change to expand the data set for training the model to improve the model's generalization ability.

[0034] In one embodiment, the step of preprocessing the deformation data and the temperature data to obtain input data includes: S24: Divide the cleaned data and the enhanced data into a training set, a validation set, and a test set according to a predetermined ratio.

[0035] Specifically, the data after cleaning and enhancement are divided into training set, validation set and test set in the ratio of 7:2:1, which are used for model parameter training and allowing the model to learn the patterns in the data; the validation set is used to adjust the model parameters during training; the test set is used to evaluate the performance of the model after training and to determine the prediction accuracy of the model on new data.

[0036] In one embodiment, the steps of combining a long short-term memory network and a convolutional neural network, initializing each model parameter, and constructing a prediction model of the hybrid neural network based on the input data include: S31: The hybrid neural network includes a convolutional neural network for feature extraction of input data and a long short-term memory network for time series analysis of the features.

[0037] Specifically, the hybrid neural network extracts features from the input data, where the features of the deformation-related data can include time-domain features, such as mean, standard deviation, kurtosis, and skewness, as well as frequency-domain features, such as the main frequency and frequency band energy distribution after Fourier transform. The features of the temperature-related data can include the temperature change rate, temperature gradient, and temperature fluctuation patterns. Based on the extracted features and combined with the robotic arm state corresponding to the data segment, the hybrid neural network is trained to enable it to analyze new input data and determine the corresponding state of the robotic arm at that time. The sources of information on the robotic arm state can include setting threshold judgments, deviation judgments, actual data collection, fault simulation experiments, etc.

[0038] In one embodiment, the prediction model uses forward propagation to output the corresponding predicted shape variable based on real-time data, including: S41: Adjust the prediction model based on the evaluation results of the loss function, where the loss function is defined by the mean square error.

[0039] The loss function can be defined using mean square error as follows:

[0040] in MSE Refers to the mean square error, x i For the i The actual deformation value of the sample, y i For the i The predicted deformation value of each sample is used to measure the deviation between the model's predicted value and the true value. The smaller the value, the closer the prediction result is to the actual value, and vice versa. The mean square error can be used to more accurately adjust the model.

[0041] In one embodiment, the prediction model uses forward propagation to output the corresponding predicted shape variable based on real-time data, including: S42: Obtain the prediction error based on the evaluation result, use the back propagation algorithm to pass the prediction error back to each layer of neurons, and update the prediction model according to the gradient descent principle.

[0042] After evaluating the prediction model using mean squared error, the backpropagation algorithm is used to update the model parameters to reduce the loss function value, thereby continuously improving the model's prediction accuracy. The performance of the prediction model is repeatedly or regularly evaluated using a validation set. Training is terminated when the loss function value of the validation set no longer decreases or overfitting occurs.

[0043] In one embodiment, the step of using a loss function to determine the real-time state of the robotic arm by predicting the deformation amount includes: S51: The actual deformation value is obtained based on the deformation data, and the loss function quantifies, compares, and evaluates the difference between the predicted deformation variable and the actual deformation variable.

[0044] The actual deformation value is converted based on the deformation data and is used in the loss function to compare with the predicted deformation value and evaluate the prediction model.

[0045] In one embodiment, the step of using a loss function to determine the real-time state of the robotic arm by predicting the deformation amount includes: S52: Combine the evaluation of the loss function with the data characteristics under different working conditions to establish the data change rules under different working conditions; S53: The prediction model dynamically selects the corresponding judgment result based on the data change law.

[0046] Different working conditions include heavy loads, material fatigue, high temperatures, etc. Data is labeled based on the working conditions, so that the prediction model can extract features from data under different working conditions, which can avoid misjudgment caused by data fluctuations due to differences in working conditions and further improve the accuracy of the prediction model.

[0047] In one embodiment, the step of using a loss function to determine the real-time state of the robotic arm by predicting the deformation amount includes: S54: When the state of the robot arm is determined to be a fault, the judgment result, fault type, fault location, fault severity, and maintenance suggestions are fed back to the operator through a visual interface.

[0048] Specifically, the robotic arm provides information feedback to the operator directly or indirectly through the host computer, including but not limited to the robotic arm's status determination results, and if a fault exists, the fault type, location, severity, and maintenance recommendations. The fault type refers to the specific fault in the robotic arm, such as mechanical faults such as loose joints, wear, fatigue, and cracks; electrical faults such as poor connections, motor stalls, and sensor anomalies; and communication faults such as program deviations and communication signal interruptions. Depending on the fault, corresponding maintenance recommendations can be made. For example, recommendations for mechanical faults may include downtime maintenance, lubrication maintenance, and in-depth testing; recommendations for electrical faults may include circuit testing and sensor calibration; and recommendations for communication faults may include program optimization, system calibration, and network configuration.

[0049] The above description is merely an embodiment of the present application and does not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A method for state detection and fault diagnosis of a robotic arm, characterized in that: Including steps: S10: Obtain deformation data and temperature data of each joint of the robotic arm through thin film sensors; S20: Preprocessing the deformation data and the temperature data to obtain input data; S30: combining the long short-term memory network and the convolutional neural network, initializing each model parameter and constructing a prediction model of the hybrid neural network based on the input data; S40: The prediction model outputs the corresponding predicted shape variable using forward propagation according to the real-time data; S50: Using a loss function to determine the real-time state of the robotic arm through the predicted deformation amount.

2. The state detection and fault diagnosis method for a robotic arm according to claim 1, characterized in that: The step of preprocessing the deformation data and the temperature data to obtain input data includes: S21: Cleaning the collected deformation data and temperature data and checking for missing values. If the missing values ​​do not exceed a threshold, data filling is performed; if the missing values ​​exceed a threshold, the corresponding data is deleted. S22: Output the cleaned data.

3. The state detection and fault diagnosis method for a robotic arm according to claim 2, characterized in that: The step of preprocessing the deformation data and the temperature data to obtain input data includes: S23: Perform data enhancement on the deformation data and the temperature data to obtain enhanced data.

4. The state detection and fault diagnosis method for a robotic arm according to claim 3, characterized in that: The step of preprocessing the deformation data and the temperature data to obtain input data includes: S24: Divide the cleaned data and the enhanced data into a training set, a validation set, and a test set according to a predetermined ratio.

5. The state detection and fault diagnosis method for a robotic arm according to claim 1, characterized in that: The steps of combining the long short-term memory network and the convolutional neural network, initializing the model parameters, and constructing a prediction model of the hybrid neural network based on the input data include: S31: The hybrid neural network includes a convolutional neural network for extracting features from the input data and a long short-term memory network for performing time series analysis on the features.

6. The state detection and fault diagnosis method for a robotic arm according to claim 1, characterized in that: The prediction model outputs the corresponding predicted shape variable using forward propagation according to the real-time data, including: S41: Adjusting the prediction model based on the evaluation result of the loss function, where the loss function is defined by the mean square error.

7. The state detection and fault diagnosis method for a robotic arm according to claim 6, characterized in that: The prediction model outputs the corresponding predicted shape variable using forward propagation according to the real-time data, including: S42: Obtain prediction errors based on the evaluation results, use a back propagation algorithm to transmit the prediction errors back to neurons in each layer, and update the prediction model according to the gradient descent principle.

8. The state detection and fault diagnosis method for a robotic arm according to claim 1, characterized in that: The step of using a loss function to determine the real-time state of the robotic arm through the predicted deformation amount includes: S51: obtaining an actual deformation value based on the deformation data, and quantifying, comparing, and evaluating the difference between the predicted deformation value and the actual deformation value using the loss function.

9. The state detection and fault diagnosis method for a robotic arm according to claim 1, characterized in that: The step of using a loss function to determine the real-time state of the robotic arm through the predicted deformation amount includes: S52: combining the evaluation of the loss function with the data characteristics under different working conditions to establish data change patterns under different working conditions; S53: The prediction model dynamically selects a corresponding determination result based on the data change rule.

10. The state detection and fault diagnosis method for a robotic arm according to claim 1, characterized in that: The step of using a loss function to determine the real-time state of the robotic arm through the predicted deformation amount includes: S54: When the state of the robotic arm is determined to be a fault, the determination result, fault type, fault location, fault severity, and maintenance suggestions are fed back to the operator through a visual interface.

Images