Multi-dimensional intelligent analysis and fault traceability system for ACU final detection test data

Abstract

The invention discloses a multi-dimensional intelligent analysis and fault traceability system for ACU final test data, and relates to the technical field of data analysis. The multi-source heterogeneous data acquisition module is used for acquiring operation parameters, function test data and assembly data through a CAN bus, test equipment and a production system, and is provided with a high-precision timestamp; the data cleaning and preprocessing module is used for standardizing data, filtering and denoising, synchronizing time and complementing missing values; the multi-dimensional feature extraction module is used for extracting features from a time domain, a frequency domain, a time sequence and a space; the intelligent fault diagnosis module is used for classifying faults by using a random forest and a D-S evidence theory and evaluating severity; the fault traceability reasoning module is used for constructing a fault propagation graph and positioning root causes through a Bayesian network; and the visual display and early warning module is used for generating a three-dimensional fault tree and a thermodynamic diagram and triggering graded early warning. According to the invention, the fault detection rate and traceability efficiency are improved, the false alarm rate is reduced, early warning is realized, the ACU test process is optimized, and the product quality and safety are guaranteed.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data analysis, and particularly relates to a multi-dimensional intelligent analysis and fault tracing system for ACU final inspection test data. BACKGROUND

[0002] With the complication of automotive electronic control systems, the analysis and fault tracing of ACU (airbag control unit) final inspection test data face severe challenges. Traditional test systems mainly rely on single-dimensional data analysis, such as focusing only on voltage, current and other electrical parameters, which is difficult to fully capture potential faults of ACU under complex working conditions. For example, intermittent faults caused by poor contact may occur in vibration tests, but traditional systems cannot effectively identify such dynamic fault patterns, which may lead to missed detection. At the same time, existing technologies lack deep fusion capabilities for multi-source heterogeneous data. Assembly data (such as bolt tightening torque), test data (such as functional response time) and historical fault data (such as specific batch failure records) generated during ACU production are usually stored in isolation and cannot be correlated for analysis, making it difficult to quickly locate the root cause during fault tracing. For example, when a batch of ACUs has high-frequency faults, it is difficult to identify tightening process deviation problems in a timely manner due to the inability to timely correlate assembly data.

[0003] Current fault diagnosis techniques are mostly based on threshold comparison or simple rule matching, which lack sensitivity to early and weak fault signals. For example, slow drift of capacitance parameters may indicate potential failure, but traditional threshold detection methods require a parameter to deviate from the standard value by more than 20% to trigger an alarm, at which point the fault may have already developed to the point of affecting functionality. In addition, existing systems have weak capabilities in fault propagation path analysis. When a component fault triggers a chain reaction, it is difficult to accurately distinguish between root cause faults and secondary faults. For example, sensor faults may cause signal abnormalities, which in turn may cause control algorithms to misjudge, ultimately resulting in actuator faults. However, traditional systems often focus directly on actuators, leading to low maintenance efficiency.

[0004] With the increasing demand for safety and reliability in the automotive industry, existing ACU test data analysis and fault tracing technologies have been unable to meet the needs. On the one hand, the popularity of new energy vehicles has brought higher electromagnetic interference risks, making ACU fault patterns more complex and diverse; on the other hand, autonomous driving technology has higher requirements for redundant design of safety systems, requiring more accurate fault diagnosis and tracing capabilities. For example, in L3 and above autonomous driving, ACU needs to have fault prediction capabilities, while existing technologies can only detect and locate faults after they occur, which cannot meet the needs of proactive maintenance. Therefore, developing a multi-dimensional intelligent analysis and fault tracing system has become a key to improving ACU quality and safety. SUMMARY

[0005] The present invention proposes a multi-dimensional intelligent analysis and fault tracing system for ACU final inspection test data to solve the problems mentioned in the prior art.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a multi-dimensional intelligent analysis and fault tracing system for ACU final inspection test data, comprising: Data acquisition module: Real-time acquisition of ACU operating parameters via CAN bus, acquisition of functional test data from test equipment, import of assembly data from production system, and all data carries a timestamp based on GPS clock. Data cleaning and preprocessing module: Standardizes data to IEEE 754 double-precision floating-point numbers, uses sliding window filtering to remove high-frequency noise, integrates multi-source data through timestamp alignment algorithm, marks missing values ​​and uses linear interpolation to complete them; Feature extraction modules for each dimension: extract mean, variance, and kurtosis statistics from the time domain; extract the amplitude of frequency components from 0-500Hz from the frequency domain using FFT; construct the Markov chain transition matrix from the time-series dimension; and calculate the Pearson correlation coefficient of sensor data from the spatial dimension. Fault diagnosis module: Based on a random forest model with 100 decision trees, it classifies fault types and uses DS evidence theory to fuse features from various dimensions. For faults with confidence scores < 0.8, it triggers a secondary test. Fault tracing and reasoning module: Constructs a directed graph of fault propagation, calculates the posterior probability of each component's fault through a Bayesian network, and uses Dijkstra's algorithm to locate the root cause component corresponding to the shortest propagation path; Visualization and early warning module: Generates a 3D fault tree and a real-time fault heat map. When parameters exceed the threshold, it triggers sound, light, SMS, and email warnings according to the severity.

[0007] Furthermore, the data cleaning and preprocessing module also includes an abnormal data identification subunit, which employs the Isolation Forest algorithm: Where S(x,n) is the anomaly score of sample x, E(h(x)) is the average path length of the sample on the isolated tree, and c(n) is the tree average path length correction function. , The Euler constant is approximately 0.5772. During implementation, 100 isolated trees are constructed. When S(x,n) > 0.7, it is marked as an anomaly and removed or corrected after manual review.

[0008] Furthermore, the fault severity assessment formula in the fault diagnosis module is as follows: Where S represents the overall severity, S func For functional impact, S safety For safety risk level, S reliabilityFor reliability impact, w1=0.5, w2=0.3, w3=0.2. During implementation, the basic score for each dimension is first determined by expert scoring, and then weighted according to the formula. When S≥8, it is judged as a serious failure.

[0009] Furthermore, the temporal feature extraction subunit of each dimension feature extraction module: uses an LSTM network to predict parameter values ​​for the next three time steps and calculates the prediction residual. The mean and maximum values ​​of the residuals are extracted as time-series features to identify parameter drift faults.

[0010] Furthermore, the root cause localization algorithm of the fault tracing and reasoning module: , where R i For component C i The root cause probability, P(C) i Let P(F|C) be the prior failure probability of the component. i Let R be the conditional probability that a failure of component Ci leads to a system failure F; during implementation, R is... i The top 3 were sorted and identified as suspicious root causes, which were then verified in conjunction with the assembly data.

[0011] Furthermore, the 3D fault tree generation algorithm of the visualization and early warning module: node radius Where P is the failure probability, the hierarchy depth corresponds to the number of failure propagation steps, and the distance between adjacent nodes is... Where k is a proportionality constant and P(F|C) is a conditional probability, interactive viewing is achieved through WebGL rendering.

[0012] Furthermore, the CAN bus data parsing subunit of the data acquisition module supports the CAN2.0A / B protocol, parses ID ranges of 0x000-0x7FF and 0x800-0x1FFF, maps the original message to physical quantities according to the DBC file, caches data every 10ms and uploads it in batches.

[0013] Furthermore, the model training subunit of the fault diagnosis module uses the Adam optimizer to train a random forest model. The training set contains historical fault data. The accuracy of the validation set is calculated every 5 epochs. Training stops when the accuracy improves by less than 0.1% for 3 consecutive epochs. The model is fine-tuned periodically with new data.

[0014] Furthermore, the Petri net analysis subunit of the fault tracing reasoning module defines places and transitions, and uses state equations... Calculate the fault propagation path, where M is the state vector, A is the correlation matrix, and t is the transition trigger vector, supporting visual modeling of 5 propagation relationships.

[0015] Furthermore, the dynamic threshold adjustment subunit of the visualization and early warning module adopts a formula. Where t is the equipment running time, T0 is the initial threshold, and k is the aging coefficient; during implementation, the threshold is updated every 24 hours to ensure the sensitivity of the early warning after the equipment ages.

[0016] Compared with existing technologies, the beneficial effects of this invention are: The system simultaneously collects multi-dimensional information such as electrical parameters, mechanical performance, and assembly data. Through intelligent algorithms, it mines the correlations between these data points, enabling it to identify complex faults and early, subtle faults that are difficult to detect using traditional methods. This significantly reduces the missed detection rate and improves the quality of ACUs leaving the factory. In terms of fault mode identification, the system constructs fault propagation models and root cause localization algorithms to accurately distinguish between primary and secondary faults, quickly pinpointing the root cause of the problem, reducing troubleshooting time, and improving maintenance efficiency.

[0017] The system's dynamic threshold adjustment mechanism and time-series feature analysis capabilities effectively solve the problem of traditional threshold detection methods being insensitive to equipment aging. By dynamically adjusting the warning threshold based on equipment operating time and extracting parameter change trend characteristics, potential faults can be predicted in advance, enabling preventative maintenance, reducing the failure rate of the ACU in actual use, and improving user safety. Furthermore, the system's 3D visualization function presents complex fault relationships in an intuitive way, allowing technicians to quickly understand the fault propagation path and impact range, assisting in decision-making, and reducing maintenance delays caused by human error.

[0018] The application of this system has also brought significant management benefits. By establishing a unified data management platform and integrating production, testing, and after-sales data, it has enabled full lifecycle data traceability and analysis, providing strong support for quality improvement. Simultaneously, the system's intelligent early warning function can monitor the testing process in real time, promptly alerting when abnormal trends are detected, preventing batch quality problems and reducing production costs. In terms of production efficiency, the automated fault diagnosis and tracing process reduces manual intervention, shortens the testing cycle, and increases production speed, giving the company a competitive edge in the fierce market. Attached Figure Description

[0019] Figure 1 This is a schematic block diagram of the multi-dimensional intelligent analysis and fault tracing system for ACU final inspection test data proposed in this invention. Figure 2 This is a schematic diagram illustrating the fault tracing probability of the multi-dimensional intelligent analysis and fault tracing system for ACU final inspection test data proposed in this invention. Figure 3 This is a schematic diagram illustrating the dynamic adjustment of the early warning threshold of the multi-dimensional intelligent analysis and fault tracing system for ACU final inspection test data proposed in this invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," and "counterclockwise," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0022] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, features defined with "first" and "second" may explicitly or implicitly include one or more of the stated features. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified. Furthermore, the terms "installed," "connected," and "linked" should be interpreted broadly; for example, they may refer to a fixed connection, a detachable connection, or an integral connection; they may refer to a mechanical connection or an electrical connection; they may refer to a direct connection or an indirect connection through an intermediate medium; and they may refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances. The invention will now be described in further detail with reference to the accompanying drawings.

[0023] Reference Figures 1 to 3 : A specific implementation method for a multi-dimensional intelligent analysis and fault tracing system for ACU final inspection test data.

[0024] I. System Architecture Deployment This application's system adopts a distributed architecture, comprising an edge layer, a network layer, and a central layer. The edge layer is deployed at the test site, configured with industrial-grade data acquisition terminals (such as Advantech UNO-2272G, supporting 4 CAN interfaces, 8 analog inputs, and 16 digital I / Os), with a sampling frequency set to 1kHz to meet high-frequency signal acquisition requirements. The network layer uses industrial Ethernet (IEEE 802.3 standard) to enable communication between edge devices and the central server, with a transmission bandwidth of 100Mbps to ensure data real-time performance. The central layer is deployed in the cloud, configured with a high-performance server cluster (CPU: Intel Xeon Silver 4210R, memory: 256GB, storage: 4TB SSD) to run data analysis and fault tracing algorithms.

[0025] II. Multi-source heterogeneous data acquisition ACU operating parameters (voltage, current, temperature, etc., sampling frequency 1kHz) are collected in real time via CAN bus (baud rate 500kbps), functional test data (braking response time, signal transmission delay, etc., accuracy ±1ms) are obtained from test equipment, and assembly data (part batch, tightening torque, etc.) are imported from the production system. All data carries a timestamp based on GPS clock (synchronization error ≤100ns). The system communicates with the ACU via a CAN interface, parsing standard DBC files (such as the SAE J1939 protocol) to acquire real-time parameters such as voltage (range ±20V, resolution 0.01V), current (range ±10A, resolution 1mA), and temperature (range -40℃~125℃, resolution 0.1℃). Simultaneously, it connects to functional testing equipment (such as a Keysight 34972A data acquisition unit) via an RS-232 interface to acquire functional test data such as braking response time (accuracy ±0.1ms) and signal transmission delay (accuracy ±0.01ms). Production assembly data (such as bolt tightening torque and assembly time) is obtained from the MES system via the OPCUA protocol, and all data is timestamped using a GPS timing module (accuracy ±100ns).

[0026] III. Data Cleaning and Preprocessing The data was standardized to IEEE 754 double-precision floating-point numbers, and high-frequency noise was removed by sliding window filtering (window size of 5 sampling points, step size of 1). Multi-source data were integrated by timestamp alignment algorithm (error ≤ 50μs), missing values ​​were marked and filled by linear interpolation. The collected data was first standardized, uniformly converted to IEEE 754 double-precision floating-point format for storage. A sliding window filtering algorithm was used to remove noise, with a window size of 5 sampling points and a step size of 1. For outlier data, the Isolation Forest algorithm was used for identification. Where S(x,n) is the anomaly score of sample x, E(h(x)) is the average path length of the sample on the isolated tree, and c(n) is the tree average path length correction function. The system constructs 100 isolated trees, with 256 samples randomly selected from each tree. Samples with S(x,n) > 0.7 are marked as anomalies. For missing values, linear interpolation is used for completion, with an interpolation interval not exceeding 10 sampling points.

[0027] IV. Multi-dimensional Feature Extraction The system extracts statistical quantities such as mean, variance, and kurtosis from the time domain, extracts the amplitude of frequency components from 0-500Hz from the frequency domain using FFT (1024 sampling points), constructs a Markov chain transition matrix from the time dimension, and calculates the Pearson correlation coefficient of sensor data from the spatial dimension. Features were extracted from four dimensions: time domain, frequency domain, time series, and spatial. Time domain features were extracted using a sliding window statistical method, calculating parameters such as mean, variance, and kurtosis within a 100ms window. Frequency domain features were obtained through FFT transformation, with 1024 sampling points and a frequency resolution of 0.977Hz (1kHz / 1024), extracting frequency component amplitudes within the 0-500Hz range. Time series features were extracted using an LSTM network with a structure of 100 input neurons (corresponding to 100ms of data), 50 hidden neurons, and 3 output neurons (predicting the next 3 time steps), calculating the statistical characteristics of the prediction residuals. Spatial features were calculated using the Pearson correlation coefficient for each sensor data point, with a threshold of 0.8; correlation coefficients below this threshold were marked as potential faults.

[0028] V. Intelligent Fault Diagnosis A random forest model based on 100 decision trees is used to classify fault types. The DS evidence theory is used to fuse multi-dimensional features (the basic probability assignment function is set based on feature importance). For faults with confidence < 0.8, a secondary test is triggered (three sets of extreme working condition tests are added). Random forest algorithm was used for fault classification. The training set contained 100,000 samples (70% normal data, 30% fault data), and the number of decision trees was set to 100. The classification results were fused using the DS evidence theory. Where m(A) is the basic probability assignment of proposition A, and K is the normalization constant. When the fused confidence level is <0.8, a secondary verification process is triggered, adding 3 sets of extreme condition tests (e.g., temperature shock test: -40℃ to 125℃, heating / cooling rate 5℃ / min). The severity of the fault is assessed using a weighted formula: Among them, S func Functional impact score (0-10 points), S safety For safety risk level (refer to ISO 26262 standard), S reliabilityThe reliability impact is determined by the MTBF decay rate. When S ≥ 8, it is considered a fatal fault, and a level 3 warning is immediately triggered.

[0029] VI. Fault Origin Reasoning Construct a directed graph of fault propagation (nodes are components, and edge weights are propagation probabilities), calculate the posterior probability of each component's fault using a Bayesian network, and use Dijkstra's algorithm to locate the root cause component corresponding to the shortest propagation path; Construct a directed graph for fault propagation, where nodes represent components (e.g., sensors, controllers, actuators), edges represent fault propagation paths, and edge weights are propagation probabilities. Use a Bayesian network to calculate the root cause probability. , where P(C i P(F|C) represents the prior fault probability (based on historical data statistics). i The conditional probability (obtained from the failure mode library) is represented by Ri. The system sorts Ri and selects the top 3 as suspected root causes. Simultaneously, a Petri net model is used to analyze the failure propagation path, defining places (component states: normal / failed) and transitions (failure propagation conditions) through state equations. Calculate the probability of fault propagation, where M is the state vector, A is the incidence matrix, and t is the transition trigger vector.

[0030] VII. Visual Display and Early Warning Generate a 3D fault tree (node ​​size is positively correlated with fault probability) and a real-time fault heat map (red indicates severe faults). When key parameters exceed the threshold (such as temperature > 85℃), trigger audible and visual (level 1), SMS (level 2), and email (level 3) warnings according to the severity. A 3D fault tree is generated using WebGL technology, with node radius... (P represents the probability of failure), and the hierarchy depth represents the number of steps in the failure propagation. The failure heatmap uses an RGB color mapping (green: normal, yellow: warning, red: severe failure), and the color depth is proportional to the severity of the failure. The warning threshold uses a dynamic adjustment mechanism. Where T0 is the initial threshold (e.g., temperature 85℃), k is the aging coefficient (taken as 0.5), and t is the equipment operating time (hours). The system updates the threshold every 24 hours to achieve accurate early warning of aging equipment. When an early warning is triggered, the first-level warning (audible and visual) is implemented through the on-site industrial display and buzzer, the second-level warning (SMS) is sent through the GSM module, and the third-level warning (email) is sent to the designated email address via the SMTP protocol.

[0031] VIII. Representation and Interpretation of Effect Data Evaluation index Traditional system This system Fault detection rate About 85% Significant improvement Fault location time Average 2 hours Significantly shortened False alarm rate About 15% Greatly reduced Early fault warning rate Less than 30% Significantly improved The above data is based on a three-month production test comparison at an automotive electronics factory. The improved fault detection rate is attributed to multi-dimensional feature extraction and intelligent fusion algorithms, which can capture complex and subtle faults that are difficult for traditional systems to identify. The reduced fault location time stems from the combined application of Bayesian and Petri nets, quickly pinpointing the root cause component and reducing the scope of investigation. The reduced false alarm rate is due to the dynamic threshold adjustment mechanism and the confidence filtering of DS evidence theory, avoiding misjudgments caused by environmental interference or parameter fluctuations. The improved early fault warning rate is attributed to LSTM time-series prediction and residual analysis, which can identify potential fault signs such as parameter drift in advance, enabling preventative maintenance. Overall, this application's system comprehensively optimizes the ACU final inspection test process through technological innovation, providing a more reliable guarantee for automotive safety.

[0032] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A multi-dimensional intelligent analysis and fault tracing system for ACU final inspection test data, characterized in that, include: Data acquisition module: Real-time acquisition of ACU operating parameters via CAN bus, acquisition of functional test data from test equipment, import of assembly data from production system, and all data carries a timestamp based on GPS clock. Data cleaning and preprocessing module: Standardizes data to IEEE 754 double-precision floating-point numbers, uses sliding window filtering to remove high-frequency noise, integrates multi-source data through timestamp alignment algorithm, marks missing values ​​and uses linear interpolation to complete them; Feature extraction modules for each dimension: extract mean, variance, and kurtosis statistics from the time domain; extract the amplitude of frequency components from 0-500Hz from the frequency domain using FFT; construct the Markov chain transition matrix from the time-series dimension; and calculate the Pearson correlation coefficient of sensor data from the spatial dimension. Fault diagnosis module: Based on a random forest model with 100 decision trees, it classifies fault types and uses DS evidence theory to fuse features from various dimensions. For faults with confidence scores < 0.8, it triggers a secondary test. Fault tracing and reasoning module: Constructs a directed graph of fault propagation, calculates the posterior probability of each component's fault through a Bayesian network, and uses Dijkstra's algorithm to locate the root cause component corresponding to the shortest propagation path; Visualization and early warning module: Generates a 3D fault tree and a real-time fault heat map. When parameters exceed the threshold, it triggers sound, light, SMS, and email warnings according to the severity.

2. The system according to claim 1, characterized in that, The data cleaning and preprocessing module also includes an abnormal data identification subunit, which uses the isolated forest algorithm. Where S(x,n) is the anomaly score of sample x, E(h(x)) is the average path length of the sample on the isolated tree, and c(n) is the tree average path length correction function. , The Euler constant is approximately 0.5772. During implementation, 100 isolated trees are constructed. When S(x,n) > 0.7, it is marked as an anomaly and removed or corrected after manual review.

3. The system according to claim 1, characterized in that, The fault severity assessment formula in the fault diagnosis module is as follows: Where S represents the overall severity, S func For functional impact, S safety For safety risk level, S reliability For reliability impact, w1=0.5, w2=0.3, w3=0.

2. During implementation, the basic score for each dimension is first determined by expert scoring, and then weighted according to the formula. When S≥8, it is judged as a serious failure.

4. The system according to claim 1, characterized in that, The temporal feature extraction subunit of the multi-dimensional feature extraction module: uses an LSTM network to predict parameter values ​​for the next three time steps and calculates the prediction residual. The mean and maximum values ​​of the residuals are extracted as time-series features to identify parameter drift faults.

5. The system according to claim 1, characterized in that, The root cause localization algorithm of the fault tracing and reasoning module: , where R i For component C i The root cause probability, P(C) i Let P(F|C) be the prior failure probability of the component. i Let R be the conditional probability that a failure of component Ci leads to a system failure F; during implementation, R is... i The top 3 were sorted and identified as suspicious root causes, which were then verified in conjunction with the assembly data.

6. The system according to claim 1, characterized in that, The 3D fault tree generation algorithm of the visualization and early warning module: node radius Where P is the failure probability, the hierarchy depth corresponds to the number of failure propagation steps, and the distance between adjacent nodes is... Where k is a proportionality constant and P(F|C) is a conditional probability, interactive viewing is achieved through WebGL rendering.

7. The system according to claim 1, characterized in that, The CAN bus data parsing subunit of the data acquisition module supports the CAN2.0A / B protocol, parses ID ranges of 0x000-0x7FF and 0x800-0x1FFF, maps the original message to physical quantities according to the DBC file, caches data every 10ms and uploads it in batches.

8. The system according to claim 1, characterized in that, The model training subunit of the fault diagnosis module uses the Adam optimizer to train a random forest model. The training set contains historical fault data. The accuracy of the validation set is calculated every 5 epochs. Training stops when the accuracy improves by less than 0.1% for 3 consecutive epochs. The model is fine-tuned periodically with new data.

9. The system according to claim 1, characterized in that, The Petri net analysis subunit of the fault tracing and reasoning module defines places and transitions, and uses state equations. Calculate the fault propagation path, where M is the state vector, A is the correlation matrix, and t is the transition trigger vector, supporting visual modeling of 5 propagation relationships.

10. The system according to claim 1, characterized in that, The dynamic threshold adjustment subunit of the visualization and early warning module uses a formula. Where t is the equipment running time, T0 is the initial threshold, and k is the aging coefficient; during implementation, the threshold is updated every 24 hours to ensure the sensitivity of the early warning after the equipment ages.

Images