A cloud-based robot fault diagnosis method

Abstract

A cloud-based robot fault diagnosis method comprises the following steps: S1, data collection and preprocessing, during the operation of the robot, by collecting robot observation data, preprocessing and fusing the data through a human-computer interaction module, and reporting to the cloud; S2, observation data storage; S3, feature extraction and fusion, feature extraction and feature fusion are performed on the robot observation data; S4, Informer model design and training, the Informer model is a deep learning model designed for processing time series data, which improves the efficiency and effect of processing time series data and is used for time series prediction tasks; S5, fault prediction and diagnosis, during the operation of the robot, real-time observation data is monitored and analyzed in real time through the Informer model; S6, Informer model evaluation and optimization; the present invention performs centralized management and sharing of observation data through cloud-based diagnosis, and utilizes continuous optimization of the model to make robot maintenance and management more efficient and intelligent.

Description

Technical Field

[0001] The present invention relates to the field of robotics technology, and in particular to a cloud-based robot fault diagnosis method. Background Art

[0002] With the rapid development of robot technology, robots have been widely used in industrial manufacturing, medical services, home assistants, military defense and other fields. The popularity of robots in these fields makes their stability and reliability particularly important. Any failure will lead to production interruption, service failure or potential safety hazards. Therefore, the research and application of fault diagnosis technology has become crucial. The current status of robot diagnosis is based on the post-firefighting mode. When a failure occurs at the customer's site, technical personnel intervene to investigate and analyze, and experimental simulation is required to restore the cause of the accident scene, resulting in many problems that cannot be reproduced and the root cause of the problem cannot be analyzed in a timely and accurate manner. Robot fault diagnosis plays an important role in improving the reliability and stability of robot systems. However, at this stage, it still faces many technical and application limitations, including challenges in data acquisition and management, complexity of fault detection, limitations of model construction, real-time and computing resource constraints, difficulty in fault reproduction and experimental simulation, and difficulty in multi-source information fusion.

[0003] In traditional robot fault diagnosis methods, relevant data are scattered across different devices and systems, and data acquisition is incomplete and untimely, making it difficult to perform effective fault diagnosis. In addition, there are technical difficulties in real-time synchronization and centralized management of data, making it impossible to effectively integrate and utilize a large amount of sensor data. In addition, the types of robot system failures are diverse and complex, and traditional fault detection methods have low diagnostic accuracy when dealing with multiple types of failures, and are unable to effectively predict and prevent failures. Traditional fault diagnosis models often rely on a large amount of labeled data during training, but in actual applications, it is difficult to obtain comprehensive labeled data, resulting in unsatisfactory model training results. When processing data in complex and dynamic environments, the model exhibits problems of low computational efficiency and poor real-time performance. Fault diagnosis requires real-time processing and analysis of a large amount of sensor data, but is limited by computing resources and processing power, making it difficult to achieve efficient and real-time fault detection. Summary of the invention

[0004] The purpose of the present invention is to provide a cloud-based robot fault diagnosis method, which can effectively solve the technical problems mentioned in the background technology.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] A cloud-based robot fault diagnosis method comprises the following steps:

[0007] S1. Data collection and preprocessing: During the operation of the robot, the robot observation data is collected, and the data is preprocessed and integrated through the human-computer interaction module and reported to the cloud;

[0008] S2, storage of observation data, storing the robot observation data collected in step S1 in a cloud database;

[0009] S3, feature extraction and fusion, feature extraction and feature fusion are performed on the robot observation data. Feature extraction includes time domain features and statistical features; time domain features include mean, standard deviation, and peak value; statistical features include statistical information within the sliding window; feature fusion, data features from different sensors are combined into a comprehensive feature vector;

[0010] Quantitatively extract and fuse the observation data stored in the database of step S2 to form different data sets, and mine the historical data to generate different data sample sets;

[0011] S4. Informer model design and training. The Informer model is a deep learning model designed for processing time series data. It improves the efficiency and effect of processing time series data and is used for time series prediction tasks.

[0012] S5, Fault prediction and diagnosis, real-time observation data is monitored and analyzed in real time through the Informer model during the operation of the robot, potential problems are identified in advance and corresponding measures are taken;

[0013] S51. Real-time data processing, inputting the latest observation data into the trained Informer model;

[0014] S52. Fault identification, comparing the prediction results with the actual observed data, detecting abnormal patterns in the data, and defining thresholds to determine whether there is a potential fault;

[0015] S53 fault prediction alarm, based on the fault detection results to generate an alarm, generate a detailed fault diagnosis report, including the fault type, occurrence time and possible causes;

[0016] S6. Informer model evaluation and optimization: continuously improve the model by evaluating the model and optimizing the model parameters to improve the prediction accuracy and model performance. The steps of model evaluation and optimization include:

[0017] S61. Evaluation indicators, using accuracy, recall, and F1 score as indicators to evaluate model performance, based on the analysis of fault records in robot observation data, analyze actual fault cases, and verify the accuracy and robustness of the model;

[0018] S62. Model optimization: adjust the model structure and parameters according to the evaluation results. Through the actual operation of the robot, collect robot observation data under different scenarios and working conditions to improve model training and form a complete data closed loop.

[0019] Preferably, in step S4, the Informer model design includes:

[0020] The encoder is used to process the input sequence data and extract high-level features of the sequence. It contains multiple layers, each of which includes a self-attention mechanism and a feedforward network.

[0021] The decoder converts the features extracted by the encoder into a predicted structure;

[0022] Time series feature processing, processing long-term dependence and periodicity characteristics in time series.

[0023] Preferably, in step S4, the Informer model training includes:

[0024] S41, data set division, divide the data into training set, validation set and test set; the training set is used to train the model; the validation set is used to adjust the model parameters and select the best model; the test set is used to evaluate the final performance of the model;

[0025] S42. Based on the Informer model design and data set partitioning, select the corresponding loss function and optimizer, compile the model, train the model using the training set data, and optimize the model parameters to improve the performance of the model;

[0026] S43, model verification, after the training is completed, the validation set data can be used for detailed evaluation, and various performance indicators can be calculated to evaluate the actual effect of the model; performance indicators include mean square error and root mean square error;

[0027] S44. Model testing: Evaluate the accuracy and generalization ability of the model on the test set. After the model training is completed, use the test set to evaluate the performance of the model and analyze the training and validation loss curves of the model to ensure that the model is not overfitting or underfitting. Generate training and validation loss curve charts and write a model performance report.

[0028] Preferably, in step S1, the steps of data collection and preprocessing include:

[0029] S11. Data classification: Observation data includes sensor data on the robot, status data of each module, and fault alarm record data; sensor data includes low-frequency sensor data of 0.1Hz~1Hz, medium-frequency sensor data of 1Hz~10Hz, and high-frequency sensor data greater than 10Hz;

[0030] S12. Data synchronization methods, including IoT communication protocol message and file synchronization methods. IoT communication protocol message synchronization is for medium and low frequency sensor data, fault records and state change data, and file synchronization is for high frequency sensor data;

[0031] S13. Data triggering methods, including timed reporting and event triggering. For IoT communication protocol messages of medium and low frequency sensor data, set a timed reporting frequency, and immediately report specific events of failure or state change. For fault event triggering of high frequency sensor data, when a fault is detected, record the changes in the time series data of high frequency sensor data and all observed data before the fault occurs.

[0032] S14. Data preprocessing and fusion, the human-computer interaction module preprocesses the observation data, reports the observation time series data in a unified format, and synchronizes it to the cloud using the Internet of Things communication protocol message.

[0033] Preferably, in step S1, the observed data includes: robot overall status information, battery management system status, electric drive controller status, charging pile control module status, robot control module status, sensor alarm information, hybrid drive module status, cleaning consumables information, robot module fault codes, sensor status information, real-time network information, and real-time motor data.

[0034] Compared with the prior art, the present invention has the following beneficial effects:

[0035] The Informer model of the present invention is a deep learning model designed for processing long-series time series data. It predicts and diagnoses faults through the Informer model, realizes faster and more accurate fault diagnosis, and thus avoids the occurrence of faults. At the same time, cloud diagnosis can conveniently carry out centralized management and sharing of data, and provide support for continuous optimization and upgrading of the model. In this way, the maintenance and management of the robot system will be more efficient and intelligent, which will help to achieve a higher level of automation and intelligence. The present invention realizes real-time synchronization and centralized management of data through real-time processing of observation information and recording of fault record information, improves the accuracy and real-time performance of fault detection, optimizes the model construction and experimental verification process, enhances the multi-source information fusion capability, forms a data closed loop, and thus improves the reliability and stability of the robot system. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 It is a flow chart of data collection and preprocessing;

[0037] Figure 2 This is the Informer model architecture diagram;

[0038] Figure 3 It is a robot fault diagnosis flow chart. DETAILED DESCRIPTION

[0039] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0040] like Figure 1-3 As shown, a cloud-based robot fault diagnosis method is characterized by comprising the following steps:

[0041] S1. Data collection and preprocessing. During the operation of the robot, the robot observation data is collected, and the data is preprocessed and integrated through the human-computer interaction module (HCM) and reported to the cloud. The specific steps of data collection and preprocessing are as follows:

[0042] S11. Data classification: Observation data includes sensor data on the robot, status data of each module, and fault alarm record data; sensor data includes low-frequency sensor data of 0.1Hz~1Hz, medium-frequency sensor data of 1Hz~10Hz, and high-frequency sensor data greater than 10Hz;

[0043] S12. Data synchronization methods, including MQTT message and file synchronization methods. MQTT message synchronization is for medium and low frequency sensor data, fault records and status change data, and file synchronization is for high frequency sensor data.

[0044] S13. Data triggering methods, including timed reporting and event triggering. For IoT communication protocol messages of medium and low frequency sensor data, set a timed reporting frequency, such as 5HZ, and immediately report specific events of failure or state change; for fault event triggering of high frequency sensor data, when a fault is detected, record the changes in the time series data of high frequency sensor data and all observation data about 5 minutes before the fault occurs;

[0045] S14. Data preprocessing and fusion: The human-computer interaction module preprocesses the observation data, reports the observation time series data in a unified format, and synchronizes it to the cloud using the Internet of Things communication protocol (MQTT) message;

[0046] The observed data includes: robot whole machine status information, gateway (CGW), battery management system (BMS) status, electric drive controller (PCU) status, charging pile control module (PCM) status, robot control module (RCM) status, chip system (SOC) status, microcontroller (MCU) status, sensor alarm information, hybrid drive module (HDM) status, cleaning consumables information, robot module fault codes, sensor status information, real-time network information, real-time motor data;

[0047] S2, storage of observation data, storing the robot observation data collected in step S1 in a cloud database;

[0048] S3, feature extraction and fusion, feature extraction and feature fusion are performed on the robot observation data. Feature extraction includes time domain features and statistical features; time domain features include mean, standard deviation, and peak value; statistical features include statistical information within the sliding window; feature fusion, data features from different sensors are combined into a comprehensive feature vector;

[0049] Quantitatively extract and fuse the observation data stored in the database of step S2 to form different data sets, and mine the historical data to generate different data sample sets, such as statistical feature data of 1 minute, 10 minutes, and 1 hour;

[0050] S4, Informer model design and training;

[0051] The Informer model is a deep learning model designed for processing time series data. It improves the efficiency and effectiveness of processing time series data. It combines an improved self-attention mechanism and other optimization strategies and is suitable for time series prediction tasks. The design of the Informer model includes:

[0052] Encoder, which is used to process input sequence data and extract high-level features of the sequence. It contains multiple layers, each of which includes a self-attention mechanism and a feedforward network;

[0053] Decoder, which converts the features extracted by the encoder into the predicted structure;

[0054] Multi-head ProbSparse Self-attention combines the multi-head self-attention mechanism and the sparse self-attention mechanism. By selectively calculating attention and reducing computational complexity, it can provide better performance and higher efficiency when processing long time series;

[0055] Masked Multi-head ProbSparse Self-attention is a self-attention mechanism for processing sequence data, which is commonly used in natural language processing (NLP) and time series data prediction tasks;

[0056] Masked uses a masking mechanism to ensure that the model can only see the current and previous sequence elements;

[0057] Sparse attention mechanism (ProbSparse) uses the sparse attention mechanism to only calculate the most important parts, reducing the amount of calculation and improving calculation efficiency;

[0058] Multi-head self-attention mechanism (Multi-head Attention) The multi-head attention mechanism enables the model to calculate attention in different subspaces in parallel and capture dependencies at different levels;

[0059] Time series feature processing, processing long-term dependence and periodicity characteristics in time series;

[0060] Informer model training includes:

[0061] S41. Dataset division: divide the data into training set, validation set and test set. Since the observation data is time series data, the observation data can be sorted in time to determine the division point. The training set is used to train the model, usually accounting for 70%-80% of the total data set. The validation set is used to adjust model parameters and select the best model, usually accounting for 10%-15% of the total data set. The test set is used to evaluate the final performance of the model, usually accounting for 10%-15% of the total data set. According to the proportion of division, select the data of the first 70%-80% time period from the beginning of the data as the training set, select the data of the next 10%-15% time period as the validation set, and select the data of the last 10%-15% time period as the test set.

[0062] S42. Model training: use the training set data to train the Informer model and optimize the model parameters;

[0063] Based on the Informer model and data set partitioning, select the appropriate loss function and optimizer, compile the model, and train the model using the training set data. During the training process, you can adjust the model parameters to improve the performance of the model.

[0064] S43. Model validation. After training is completed, the validation set data can be used to evaluate the model in detail and calculate various performance indicators to evaluate the actual effect of the model. For example, indicators such as mean square error (MSE) and root mean square error (RMSE) can be used to measure the prediction accuracy of the model.

[0065] S44. Model testing: evaluate the accuracy and generalization ability of the model on the test set. After the model training is completed, use the test set to evaluate the performance of the model and analyze the training and validation loss curves of the model to ensure that the model is not overfitting or underfitting. Generate training and validation loss curve charts and write a model performance report;

[0066] S5. Fault prediction and diagnosis;

[0067] The process of combining real-time observation data with the Informer model to monitor and analyze the robot in real time during operation, identify potential problems in advance and take corresponding measures;

[0068] S51. Real-time data processing, inputting the latest observation data into the trained Informer model;

[0069] S52. Fault identification, comparing the prediction results with the actual observed data, detecting abnormal patterns in the data, and defining thresholds to determine whether there is a potential fault;

[0070] S53 fault prediction alarm, based on the fault detection results to generate an alarm, generate a detailed fault diagnosis report, including the fault type, occurrence time and possible causes;

[0071] S6. Model evaluation and optimization: Continuously improve the model by evaluating the model and optimizing the model parameters to improve the prediction accuracy and model performance. The steps of model evaluation and optimization include:

[0072] S61. Evaluation indicators, use accuracy, recall, F1 score and other indicators to evaluate model performance, analyze actual failure cases based on failure record analysis in robot observation data, and verify the accuracy and robustness of the model;

[0073] S62. Model optimization and model adjustment. Adjust the model structure and hyperparameters according to the evaluation results. Through the actual operation of the robot, collect robot observation data under different scenarios and working conditions to improve model training and form a complete data closed loop.

[0074] Cloud fault diagnosis is an effective solution to improve the operating efficiency of robots, reduce downtime, reduce maintenance costs, and improve the stability and reliability of the overall automation system. The Informer model of the present invention is a deep learning model designed for processing long-sequence time series data. It predicts and diagnoses faults through the Informer model to achieve faster and more accurate fault diagnosis, thereby avoiding the occurrence of faults. At the same time, cloud diagnosis can conveniently centrally manage and share data, and provide support for the continuous optimization and upgrading of the model. In this way, the maintenance and management of the robot system will be more efficient and intelligent, which will help achieve a higher level of automation and intelligence.

[0075] The present invention realizes real-time synchronization and centralized management of data through real-time processing of observation quantity information and recording of fault record information, improves the accuracy and real-time performance of fault detection, optimizes the model construction and experimental verification process, enhances the multi-source information fusion capability, forms a data closed loop, and thus improves the reliability and stability of the robot system.

[0076] The above contents are merely examples and explanations of the structure of the present invention. The technicians in this technical field may make various modifications or additions to the specific embodiments described or replace them in a similar manner. As long as they do not deviate from the structure of the present invention or exceed the scope defined by the claims, they should all fall within the protection scope of the present invention.

Claims

1. A cloud-based robot fault diagnosis method, characterized in that: The following steps are involved: S1. Data collection and preprocessing: During the operation of the robot, the robot observation data is collected, and the data is preprocessed and integrated through the human-computer interaction module and reported to the cloud; S2, storage of observation data, storing the robot observation data collected in step S1 in a cloud database; S3, feature extraction and fusion, feature extraction and feature fusion are performed on the robot observation data. Feature extraction includes time domain features and statistical features; Time domain features include mean, standard deviation, and peak value; Statistical features include statistical information within the sliding window; Feature fusion, combining data features from different sensors into a comprehensive feature vector; Quantitatively extract and fuse the observation data stored in the database of step S2 to form different data sets, and mine the historical data to generate different data sample sets; S4. Informer model design and training. The Informer model is a deep learning model designed for processing time series data. It improves the efficiency and effect of processing time series data and is used for time series prediction tasks. S5, Fault prediction and diagnosis, real-time observation data is monitored and analyzed in real time through the Informer model during the operation of the robot, potential problems are identified in advance and corresponding measures are taken; S51. Real-time data processing, inputting the latest observation data into the trained Informer model; S52. Fault identification, comparing the prediction results with the actual observed data, detecting abnormal patterns in the data, and defining thresholds to determine whether there is a potential fault; S53. Fault prediction alarm, generates alarms based on fault detection results, and generates detailed fault diagnosis reports, including fault type, occurrence time and possible causes; S6. Informer model evaluation and optimization: continuously improve the model by evaluating the model and optimizing the model parameters to improve the prediction accuracy and model performance. The steps of model evaluation and optimization include: S61. Evaluation indicators, use accuracy, recall, and F1 score as indicators to evaluate model performance, analyze actual failure cases based on failure record analysis in robot observation data, and verify the accuracy and robustness of the model; S62. Model optimization: adjust the model structure and parameters according to the evaluation results, collect robot observation data under different scenarios and working conditions through the actual operation of the robot, so as to improve model training and form a complete data closed loop; In step S1, the steps of data collection and preprocessing include: S11. Data classification: Observation data includes sensor data on the robot, status data of each module, and fault alarm record data; sensor data includes low-frequency sensor data of 0.1 Hz to 1 Hz, medium-frequency sensor data of 1 Hz to 10 Hz, and high-frequency sensor data greater than 10 Hz; S12. Data synchronization methods, including IoT communication protocol message and file synchronization methods. IoT communication protocol message synchronization is for medium and low frequency sensor data, fault records and state change data, and file synchronization is for high frequency sensor data; S13. Data triggering methods, including timed reporting and event triggering. For IoT communication protocol messages of medium and low frequency sensor data, set a timed reporting frequency, and immediately report specific events of failure or state change; for fault event triggering of high frequency sensor data, record the changes in the time series data of high frequency sensor data and all observed data before the failure occurs when a fault is detected; S14. Data preprocessing and fusion: The human-computer interaction module preprocesses the observation data, reports the observation time series data in a unified format, and synchronizes it to the cloud using the Internet of Things communication protocol message.

2. The cloud-based robot fault diagnosis method according to claim 1, characterized in that: In step S4, the Informer model design includes: The encoder is used to process the input sequence data and extract high-level features of the sequence. It contains multiple layers, each of which includes a self-attention mechanism and a feedforward network. The decoder converts the features extracted by the encoder into a predicted structure; Time series feature processing, processing long-term dependence and periodicity characteristics in time series.

3. The cloud-based robot fault diagnosis method according to claim 2, characterized in that: In step S4, the Informer model training includes: S41, data set division, divide the data into training set, validation set and test set; the training set is used to train the model; the validation set is used to adjust the model parameters and select the best model; the test set is used to evaluate the final performance of the model; S42. Based on the Informer model design and data set partitioning, select the corresponding loss function and optimizer, compile the model, train the model using the training set data, and optimize the model parameters to improve the performance of the model; S43, model verification, after the training is completed, the validation set data can be used for detailed evaluation, and various performance indicators can be calculated to evaluate the actual effect of the model; performance indicators include mean square error and root mean square error; S44. Model testing: Evaluate the accuracy and generalization ability of the model on the test set. After the model training is completed, use the test set to evaluate the performance of the model, analyze the training and validation loss curves of the model to ensure that the model is not overfitting or underfitting, generate training and validation loss curve charts, and write a model performance report.

4. The cloud-based robot fault diagnosis method according to claim 1, characterized in that: In step S1, the observed data includes: robot status information, battery management system status, electric drive controller status, charging pile control module status, robot control module status, sensor alarm information, hybrid drive module status, cleaning consumables information, robot module fault codes, sensor status information, real-time network information, and real-time motor data.

Images