Equipment fault prediction method and system based on PLC data

Abstract

The invention discloses an equipment fault prediction method and system based on PLC data, and aims to solve the problems that existing equipment fault prediction depends on manual analysis, relevance recognition is inaccurate, and early warning lags behind. The method comprises the steps that equipment operation data and sensor data in a PLC are acquired through a data acquisition module; performing feature extraction on the data by using a preset convolution kernel to obtain a multi-dimensional feature value, determining a task, a state and sensor data associated with the target fault point location based on the feature value through an association analysis module, and performing trend prediction on the associated data by using a time sequence prediction algorithm; judging the occurrence probability and the possible occurrence time of the target fault in the future preset time according to the prediction result; according to the invention, through automatic feature extraction and correlation analysis and in combination with time sequence prediction, early warning of equipment faults is realized, the prediction accuracy and timeliness are improved, and the method is suitable for intelligent maintenance of various industrial equipment.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of PLC data processing, and in particular to a device fault prediction method and system based on PLC data. BACKGROUND

[0002] Stable operation of industrial equipment is the key to production efficiency. Traditional fault handling methods are mostly "after-maintenance" or "periodic maintenance", which have the following defects: After-maintenance results in long downtime and large production losses; Periodic maintenance relies on experience and is prone to over-maintenance (waste of cost) or insufficient maintenance (fault still occurs); Existing fault prediction methods mostly rely on single sensor data, without combining global data such as device task execution and state switching, resulting in inaccurate correlation identification and high false alarm rate; Manual data analysis is low in efficiency and difficult to handle the massive time series data generated by PLC, making it impossible to achieve real-time early warning.

[0003] For example, Chinese patent CN202210567890.1 discloses a device fault early warning method based on a vibration sensor, but it only relies on vibration data and does not consider key information such as the type of task performed by the device and the load state. When the same vibration value has different fault risks under different tasks, misjudgment is likely to occur.

[0004] Therefore, there is an urgent need for a technical solution that can integrate PLC multi-dimensional data, automatically identify correlations and accurately predict faults. SUMMARY

[0005] The present application aims to solve the problems of single data dimension, inaccurate correlation identification and delayed early warning in the prior art, and provides a fault prediction method and system based on PLC full data, combining convolution feature extraction and time series prediction.

[0006] The present application is implemented by the following technical solutions: By collecting multi-dimensional data (task, state, fault, sensor) in PLC, using convolution kernel to extract data features, identifying key data related to faults, and then using time series prediction to predict the possibility of fault occurrence in advance, the following is the system composition: Data acquisition module: communicate with PLC through OPC UA protocol, acquire data in real time and store it in database (such as MySQL), sampling frequency is consistent with PLC (1-100Hz adjustable), data includes: Task execution data: task number, start time, end time, execution parameters (such as speed, pressure set value); Device status data: running mode (automatic / manual), load rate (0-100%), motor temperature (°C), valve switch status; Fault point data: fault occurrence time, point number (e.g. "motor overload protection point", "valve sticking detection point"), fault code; Sensor data: real-time values of vibration sensor (mm / s), pressure sensor (MPa), and flow sensor (m³ / h).

[0007] Feature extraction module: built-in pre-trained 1D and 2D convolution kernels, where 1D convolution kernels (e.g. [1, -1, 1]) are used to extract time series features of single-dimensional data (e.g. the change slope of sensor values within 10 seconds), and 2D convolution kernels (e.g. 3x3 matrix) are used to extract correlation features of multi-dimensional data (e.g. the combined feature of "task A execution + pressure > 5MPa + vibration > 2mm / s").

[0008] Correlation analysis module: based on the feature extraction results, calculate the correlation degree of each data with the target fault point through the grey correlation degree algorithm. For example, for the "valve sticking" fault, calculate the correlation degree of "valve switch instruction", "flow sensor value", "motor load rate", etc. with the fault, and select the key data with correlation degree ≥0.7.

[0009] Time series prediction module: use LSTM network to predict the correlation data. Take the historical 24-hour correlation data as input (time step 5 minutes), train the model to predict the data trend in the next 1-24 hours. For example, the prediction of "flow sensor value" can get the change curve in the next 12 hours.

[0010] Fault warning module: compare the predicted value with the preset threshold (e.g. "flow sensor value <0.5m³ / h" is the valve sticking warning threshold), if the predicted value exceeds the threshold within the next T hours, output the warning: "valve sticking fault (point number V3) may occur in the next 2-3 hours, probability 78%", with the trend chart of the correlation data. Advantages

[0011] Comprehensive data dimension: integrate the task, status, fault, and sensor data in PLC, avoid misjudgment caused by single data; Accurate correlation identification: through convolution feature extraction and quantitative correlation degree calculation, automatically locate the key data related to the fault, and the correlation identification accuracy is improved by more than 30%; High prediction lead time: based on time series algorithm, realize fault prediction in the next 1-24 hours, the warning lead time is increased by 2-10 times compared with traditional methods; High degree of automation: no manual intervention is required throughout the entire process, suitable for real-time monitoring needs in industrial scenarios. DETAILED DESCRIPTION

[0012] To make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with examples.

[0013] Taking the fault prediction of a certain production line's disassembly and stacking robot as an example, the implementation process of the present application is described in detail: Step 1: Data collection The data collection module communicates with the robot PLC through the OPC UA protocol and collects the following data (sampling frequency 10Hz): Task execution data: task ID (such as W101), start time (2023-10-01 08:00:00), working current (150A); Device status data: operating mode (automatic), arm temperature (55℃), servo motor load rate (60%); Fault point data: historical fault record (such as 2023-09-28 14:30:00, point M10 (servo motor overload), fault code E01); Sensor data: arm vibration (1.2mm / s), grabbing pressure (0.6MPa).

[0014] Step 2: Feature extraction The feature extraction module calls a 2D convolution kernel (3x3) to process the collected data and extract feature values: Time sequence feature: "When task W101 is executed, the vibration value rises from 1.0mm / s to 1.5mm / s within 5 seconds"; Correlation feature: "When load rate >70% and pressure >0.7MPa, the occurrence frequency of M10 point failure within 10 minutes is 80%".

[0015] Step 3: Correlation analysis Calculate the correlation degree of each data with M10 (servo motor overload): Load rate: correlation degree 0.82 (highest); Vibration value: correlation degree 0.75; Working current: correlation degree 0.68; Filter out key data with correlation degree ≥0.7: load rate, vibration value.

[0016] Step 4: Time sequence prediction Using the LSTM network, input the historical 24-hour load rate and vibration value data (time step 5 minutes, a total of 288 data points), and after training, predict the data for the next 12 hours: Prediction result: In the next 3-4 hours, the load rate will rise from 60% to 75%, and the vibration value will rise from 1.2mm / s to 1.8mm / s.

[0017] Step 5: Fault warning The preset threshold is: when the load rate > 70% and the vibration value > 1.6mm / s, the warning is triggered.

[0018] After comparing the predicted values, the fault warning module outputs: "Servo motor overload fault (point M10) may occur in the next 3-4 hours, probability 82%", and displays the predicted trend graph of load rate and vibration value on the monitoring interface.

[0019] Variation of the specific implementation For high sampling frequency (such as 100Hz) sensor data, the convolution kernel size can be adjusted to 5x5 to improve feature extraction accuracy; The time series prediction algorithm can be replaced by GRU, suitable for scenarios with limited computing resources, with a 20% increase in prediction speed; The correlation threshold can be dynamically adjusted according to the importance of the device (such as 0.8 for critical devices and 0.6 for general devices).

[0020] The device fault prediction method and system based on PLC data proposed in the present application addresses the core pain points of "single data dimension, fuzzy correlation relationship identification, and delayed warning" in existing industrial device fault prediction, and builds a "multi-source data acquisition - convolution feature extraction - correlation relationship positioning - time series trend prediction - intelligent fault warning" full-process technical system. This technology integrates device task execution data, state data, fault point data, and sensor real-time data in PLC through industrial protocols, breaking through the limitations of traditional prediction methods relying on single data; with the collaborative operation of 1D and 2D convolution kernels, it realizes the accurate extraction of data time series features and multi-dimensional correlation features, solving the problem of difficult data feature capture in complex industrial scenarios; by quantifying the correlation degree analysis to filter out key factors affecting faults, combined with LSTM / GRU time series prediction algorithms, the fault warning is advanced to 1-24 hours, significantly improving the timeliness and accuracy of warning (correlation identification accuracy improved by more than 30%, warning advanced by 2-10 times). From the application value, the system does not need to make hardware transformation to the existing PLC equipment, can be deployed only through software module integration, is suitable for various industrial scenes such as manufacturing production lines, automatic welding equipment, intelligent machine tools, can reduce equipment downtime loss and maintenance cost, can provide technical support for intelligent operation and maintenance of industrial equipment, and has wide popularization prospect and practical value.

[0021] Meanwhile, the design of dynamically adjusting the convolution kernel size and the correlation threshold further enhances the flexibility and adaptability of the technology, and can meet the fault prediction needs of different industries and different important level equipment, and provides a reliable technical scheme for the 'predictive maintenance' in the industrial field.

[0022] The above describes only the preferred specific embodiments of the present application, but the protection scope of the present application is not limited to this, any person skilled in the art can easily think of changes or replacements within the technical range disclosed by the present application, which should be covered within the protection scope of the present application.

Claims

1. A PLC data-based equipment fault prediction system, characterized in that: Includes the following modules: The data acquisition module is used to collect equipment operation data and sensor data from the PLC in real time. The equipment operation data includes task execution data, equipment status data and fault location data. The fault location data records the time of the fault and the fault status of the corresponding location. The feature extraction module has at least one preset convolution kernel built in, which is used to perform convolution operations on the device operation data and sensor data to extract multi-dimensional feature values, including data time series features, mutation features and correlation features. The correlation analysis module, connected to the feature extraction module, is used to calculate the correlation between task execution data, equipment status data, sensor data and target fault location based on the multidimensional feature values, and to filter out the correlation data with a correlation higher than a preset threshold. The time series prediction module is connected to the correlation analysis module and is used to perform trend prediction on the correlation data using a time series prediction algorithm, and output the predicted value within a preset time period in the future. The fault early warning module, connected to the time-series prediction module, is used to compare the predicted value with a preset fault threshold, determine the possible time and probability of the target fault, and output early warning information.

2. The equipment fault prediction system based on PLC data according to claim 1, characterized in that: The preset convolution kernel includes a 1D convolution kernel and a 2D convolution kernel. The 1D convolution kernel is used to extract the temporal features of the data, and the 2D convolution kernel is used to extract the correlation features between multi-dimensional data.

3. The equipment fault prediction system based on PLC data according to claim 1, characterized in that: The correlation analysis module calculates the correlation degree using Pearson correlation coefficient, mutual information, or grey relational degree algorithm, with the preset threshold being 0.6-0.

8.

4. The equipment fault prediction system based on PLC data according to claim 1, characterized in that: The temporal prediction algorithm includes a long short-term memory network, a gated recurrent unit, and a temporal convolutional network.

5. The equipment fault prediction system based on PLC data according to claim 1, characterized in that: The early warning information output by the fault early warning module includes the fault location number, the predicted occurrence time window, the occurrence probability, and an anomaly trend chart of related data.

6. A method for predicting equipment faults based on PLC data, characterized in that: Includes the following steps: S1. Data Acquisition: The data acquisition module obtains equipment operation data and sensor data from the PLC. The equipment operation data includes task execution data, equipment status data, and fault location data. The task execution data includes task ID, start and end time, and execution parameters. The equipment status data includes running, stopping, and standby status, load rate, and temperature. The fault location data includes fault occurrence time, location number, and fault code. The sensor data includes real-time sampling values ​​from vibration, pressure, and flow sensors. S2. Feature Extraction: The feature extraction module calls a preset convolution kernel to perform convolution operation on the data collected in step S1 and extract feature values. The feature values ​​include: abrupt change features of data in the time dimension, temporal features of task execution and state switching, and collaborative features among multiple data. S3. Correlation Analysis: Based on the feature values ​​from step S2, the correlation analysis module calculates the correlation degree between each task execution data, equipment status data, sensor data and the target fault location, and filters out the correlated data with a correlation degree ≥ a preset threshold. S4. Time Series Forecasting: The time series forecasting module uses a time series forecasting algorithm to perform trend forecasting on the related data selected in step S3, and obtains the predicted value within the future time period T1-T2, where T1 is the forecast start time, T2 is the forecast end time, and T2-T1 ≥ 1 hour; S5. Fault warning: The fault warning module compares the predicted value in step S4 with the preset fault threshold. If the predicted value exceeds the threshold, a warning message is output. The warning message includes the target fault location, the possible time, and the probability of occurrence.

7. The equipment fault prediction method based on PLC data according to claim 6, characterized in that: In step S2, the size of the convolution kernel is dynamically adjusted according to the data sampling frequency: when the sampling frequency is 1Hz-10Hz, the convolution kernel size is set to 3×3; when the sampling frequency is >10Hz, the convolution kernel size is set to 5×5.

8. The equipment fault prediction method based on PLC data according to claim 6, characterized in that: In step S3, the correlation degree is calculated as follows: The correlation degree is calculated as (feature similarity × 0.4 + temporal co-occurrence frequency × 0.3 + data change rate before and after the fault × 0.3), where feature similarity is calculated using cosine similarity, and temporal co-occurrence frequency is the percentage of times that the associated data and the fault location appear in the same time window.

9. The equipment fault prediction method based on PLC data according to claim 6, characterized in that: In step S4, the time series prediction algorithm uses an LSTM network, whose input is the historical time series sequence of the associated data, and whose output is the prediction sequence for the next 1-24 hours, with a prediction accuracy error ≤5%.

10. The equipment fault prediction method based on PLC data according to claim 6, characterized in that: In step S5, the preset fault threshold is determined by collecting the maximum value of the associated data from the hour before the historical fault occurred, and taking 1.2 times its average value as the threshold.