A multi-dimensional sensing-based intelligent fault diagnosis and early warning system for electromechanical equipment

Abstract

The application relates to a kind of intelligent fault diagnosis and early warning system of electromechanical equipment based on multidimensional sensing applied to the technical field of fault diagnosis, comprising: multidimensional sensing data acquisition module, for collecting multi-source time series sensing data;Data quality evaluation module, for generating quality evaluation vector, its built-in space-time uncertainty propagation unit is used to estimate the influence of data quality decline;Multi-level diagnosis engine contains the first, second diagnosis unit and diagnosis fusion unit of execution hierarchical confidence propagation, for generating comprehensive fault probability;Dynamic hierarchical warning module contains dynamic threshold calculation unit and early warning generation unit, for calculating the dynamic threshold of fusion quality evaluation vector, confidence and influence estimation result and output hierarchical early warning signal;Closed-loop feedback unit is used to correct threshold parameter according to actual fault confirmation result.The application realizes high reliability fault diagnosis and early warning under the condition of data quality uncertainty, and can effectively reduce false alarm rate.

Description

Technical Field

[0001] This invention relates to the field of fault diagnosis technology, specifically to an intelligent fault diagnosis and early warning system for electromechanical equipment based on multi-dimensional sensing, which is particularly suitable for achieving highly reliable fault diagnosis and dynamic early warning under complex operating conditions such as missing sensor data, noise, and uncertain quality. Background Technology

[0002] Electromechanical equipment (such as electric motors, gearboxes, compressors, and fans) is the core equipment of modern industrial production, and its operating status directly affects production safety and economic benefits. With the development of industrial Internet of Things and artificial intelligence technologies, intelligent fault diagnosis and early warning systems based on multi-source sensor data have become a research hotspot.

[0003] Chinese patent application CN119758923A discloses a device monitoring and prediction method and system based on industrial middleware and edge computing. This system uses a monitoring and acquisition unit to collect vibration, temperature, and current data of the equipment in real time, and analyzes the data through a three-layer prediction architecture: a lightweight threshold rule model is deployed at the edge layer for real-time anomaly detection; an XGBoost model is deployed at the middleware layer for trend analysis; and a temporal convolutional network model is deployed at the cloud layer for deep time-series analysis, establishing a corresponding hierarchical early warning mechanism. This solution achieves comprehensive monitoring of equipment through the coordinated operation of the three-layer architecture, achieving beneficial results in reducing system response latency and improving early warning accuracy.

[0004] However, this scheme has the following shortcomings: its dynamic warning threshold is mainly calculated based on statistical rules or the output of a single model, without considering the quality differences of multi-source sensor data and their propagation effects in the spatiotemporal dimensions. In real-world industrial scenarios, sensor data inevitably suffers from quality issues such as missing data, noise, and drift, which directly affect the reliability of diagnostic results. When the data quality of a certain sensor node deteriorates, existing technologies fail to assess the impact of this quality deterioration on the diagnostic confidence of adjacent sensor nodes and subsequent time steps, nor do they quantify this impact as a basis for adjusting the dynamic warning threshold.

[0005] Furthermore, in existing technologies, diagnostic confidence is usually calculated at the final output layer, without establishing a hierarchical propagation and correction mechanism for confidence within the diagnostic network, resulting in insufficient granularity in confidence assessment.

[0006] To address the aforementioned technical problems, this invention provides an intelligent fault diagnosis and early warning system for electromechanical equipment based on multi-dimensional sensing. Summary of the Invention

[0007] This invention addresses the technical problems in existing technologies, such as the failure to consider the quality differences and spatiotemporal propagation effects of multi-source sensor data in dynamic early warning thresholds and the lack of a hierarchical propagation mechanism for diagnostic confidence. It provides an intelligent fault diagnosis and early warning system for electromechanical equipment based on multi-dimensional sensing, aiming to achieve highly reliable fault diagnosis and dynamic early warning under complex operating conditions such as missing sensor data, noise, and uncertain quality.

[0008] To achieve the above objectives, the present invention provides the following technical solution: An intelligent fault diagnosis and early warning system for electromechanical equipment based on multi-dimensional sensing includes: The multi-dimensional sensor data acquisition module is used to acquire multi-source time-series sensor data from multiple heterogeneous sensors on electromechanical equipment. Heterogeneous sensors include, but are not limited to, vibration sensors, temperature sensors, current sensors, voltage sensors, acoustic emission sensors, etc.

[0009] The data quality assessment module, connected to the multi-dimensional sensor data acquisition module, is used to perform quality analysis on multi-source time-series sensor data and generate quality evaluation vectors. The data quality assessment module also includes a spatiotemporal uncertainty propagation unit, which, based on the topology and temporal correlation of the sensor network, estimates the impact of data quality degradation on the diagnostic confidence of adjacent sensor nodes and adjacent time steps, and outputs the estimated impact results.

[0010] A multi-level diagnostic engine, connected to the data quality assessment module, includes: The first diagnostic unit is used to output a first fault probability distribution and a first confidence level based on multi-source time-series sensor data. The second diagnostic unit is used to output a second fault probability distribution and a second confidence level based on multi-source time-series sensor data. The diagnostic fusion unit is used to fuse the first fault probability distribution and the second fault probability distribution to generate a comprehensive fault probability. The first diagnostic unit and the second diagnostic unit are each internally configured to perform hierarchical confidence propagation and output the first confidence level and the second confidence level after hierarchical confidence propagation correction.

[0011] The dynamic hierarchical early warning module, connected to the data quality assessment module and the multi-level diagnostic engine, includes: The dynamic threshold calculation unit is used to calculate the dynamic early warning threshold based on the quality evaluation vector, the weighted value of the first confidence level and the second confidence level, and the impact degree prediction result. The early warning generation unit is used to generate a graded early warning signal when the overall failure probability exceeds the dynamic early warning threshold.

[0012] The closed-loop feedback unit is used to correct the threshold adjustment parameters of the dynamic threshold calculation unit based on the actual fault confirmation results after the warning is triggered, so as to form an adaptive closed-loop adjustment of the warning threshold.

[0013] In the preferred technical solution, the quality evaluation vector includes at least the data integrity index and the signal-to-noise ratio index.

[0014] In the preferred technical solution, the first diagnostic unit extracts global structural features of multi-source time-series sensor data based on a first type of feature extraction algorithm to output a first fault probability distribution; the second diagnostic unit extracts local time-series features of multi-source time-series sensor data based on a second type of feature extraction algorithm to output a second fault probability distribution.

[0015] In the preferred technical solution, the dynamic threshold calculation unit is further configured to: automatically increase the dynamic warning threshold to reduce the warning sensitivity when at least one component in the quality evaluation vector is lower than the preset lower limit threshold; and send a forced re-diagnosis command to the multi-level diagnostic engine when the first confidence level or the second confidence level is lower than the preset confidence level lower limit.

[0016] In the preferred technical solution, the multi-level diagnostic engine also includes a confidence cross-validation unit, which is connected between the first diagnostic unit, the second diagnostic unit and the diagnostic fusion unit. It is used to calculate the similarity between the first fault probability distribution and the second fault probability distribution. When the similarity is lower than the preset consistency threshold, it is determined that the current diagnostic result has uncertainty, and the uncertainty identifier is used as a penalty factor when the diagnostic fusion unit performs weighted fusion to reduce the output value of the comprehensive fault probability.

[0017] In the preferred technical solution, the multi-level diagnostic engine further includes a diagnostic result credibility verification unit, connected between the diagnostic fusion unit and the dynamic hierarchical early warning module. This unit assesses the credibility of the current comprehensive diagnostic result output by the diagnostic fusion unit based on historical fault confirmation results accumulated in the closed-loop feedback unit, and outputs a credibility correction coefficient. Historical fault confirmation results include actual fault confirmation results after each early warning trigger and operational data confirming no faults under normal operating conditions. The credibility verification unit is configured to: reduce the output value of the credibility correction coefficient when the deviation between the fault type of the current comprehensive diagnostic result and the fault type distribution in the historical fault confirmation results exceeds a preset deviation threshold; and send a re-diagnosis trigger command to the multi-level diagnostic engine when the credibility correction coefficient is lower than a preset lower credibility threshold, and input the credibility correction coefficient as a penalty factor to the diagnostic fusion unit for dynamic correction of the fusion weights of subsequent diagnostic results.

[0018] In the preferred technical solution, the multi-level diagnostic engine is configured to support a progressive diagnostic mode: when the dynamic hierarchical early warning module determines that the overall fault probability exceeds the preset first early warning threshold but does not exceed the dynamic early warning threshold, the intermediate level diagnosis of the multi-level diagnostic engine is triggered. The computational complexity of the intermediate level diagnosis is lower than that of the complete diagnosis but higher than that of the independent diagnosis of the first or second diagnostic unit; when the intermediate fault probability output by the intermediate level diagnosis still exceeds the first early warning threshold, the complete diagnosis is triggered again.

[0019] In the preferred technical solution, the diagnostic fusion unit is configured to: when the difference between the first fault probability distribution and the second fault probability distribution exceeds a preset difference threshold, activate the weight adaptive adjustment mechanism, and dynamically optimize the fusion weight of the first diagnostic unit and the second diagnostic unit by minimizing the deviation between the weighted fusion result and the matching mode in the historical fault mode library.

[0020] In the preferred technical solution, the closed-loop feedback unit further includes a model optimization subunit, which uses the complete fault cycle data collected after each warning is triggered as training samples to periodically update the diagnostic models of the first diagnostic unit and the second diagnostic unit, and feeds back the updated model parameters to the dynamic threshold calculation unit to adjust the base threshold of the dynamic warning threshold.

[0021] In the preferred technical solution, the multi-level diagnostic engine also includes an abnormal data interception and uploading unit, which is used to intercept sensor data segments of a set time length before and after the triggering time when the overall fault probability exceeds a preset threshold, or when the probability of any fault type in the fault probability distribution output by the first diagnostic unit or the second diagnostic unit exceeds a preset fault identification threshold, and upload them as high-value sample data to the cloud or storage unit for subsequent optimization and updates of the diagnostic model.

[0022] Compared with the prior art, the present invention has the following beneficial effects: 1. By introducing a spatiotemporal uncertainty propagation unit into the fault diagnosis and early warning system, the impact of data quality degradation on the diagnostic confidence of adjacent nodes and adjacent time steps is quantitatively evaluated based on the sensor network topology and time correlation. The impact results are directly used for dynamic early warning threshold calculation, which significantly improves the robustness of the system in scenarios of local sensor failure or data quality fluctuation.

[0023] 2. By performing layer-by-layer propagation and weighted update of local confidence along the diagnostic network hierarchy within the diagnostic unit, the evaluation of diagnostic confidence is no longer limited to the final output layer, but incorporates the feature fusion weights of intermediate layers, thereby improving the granularity and accuracy of confidence evaluation.

[0024] 3. The dynamic early warning threshold integrates the quality evaluation vector, the diagnostic confidence weighted value, and the spatiotemporal uncertainty propagation results, realizing the adaptive adjustment of early warning sensitivity to data quality and diagnostic reliability, effectively reducing the false alarm rate and the missed alarm rate.

[0025] 4. The actual fault confirmation results are used for threshold parameter correction through the closed-loop feedback unit. Combined with the incremental update of the diagnostic model by the model optimization subunit and the active collection of high-value samples by the abnormal data interception and uploading unit, a full-chain adaptive closed loop of "diagnosis-early warning-feedback-optimization" is formed. The system performance can be continuously improved as the running data accumulates. Attached Figure Description

[0026] Figure 1 This is a schematic diagram of the overall architecture of the intelligent fault diagnosis and early warning system for electromechanical equipment based on multi-dimensional sensing in an embodiment of the present invention. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of protection of this invention.

[0028] Example 1: System Overall Architecture like Figure 1 As shown, this embodiment provides an intelligent fault diagnosis and early warning system for electromechanical equipment based on multi-dimensional sensing, including: a multi-dimensional sensing data acquisition module, a data quality assessment module, a multi-level diagnostic engine, a dynamic hierarchical early warning module, and a closed-loop feedback unit.

[0029] The multi-dimensional sensor data acquisition module is deployed at key measurement points of electromechanical equipment to collect multi-source time-series sensor data in real time from multiple heterogeneous sensors (e.g., acceleration vibration sensors mounted on motor bearing housings, temperature sensors mounted on stator windings, and current sensors mounted on power lines), including vibration signals, temperature signals, and current signals. The acquired data is timestamped and preliminarily filtered before being transmitted to the data quality assessment module.

[0030] The data quality assessment module receives multi-source time-series sensor data and first performs quality analysis on the data from each sensor channel. Specifically, the data quality assessment module calculates a data integrity index (e.g., the proportion of valid data to total sampling time) and a signal-to-noise ratio (SNR) index (e.g., the ratio of signal power to noise power) for each channel. It should be noted that the quality evaluation vector can be a combination of one or more indicators that characterize data quality, including but not limited to data integrity index, SNR index, root mean square error, peak signal-to-noise ratio, etc. This embodiment uses the integrity index and SNR index as examples. The indicators of each channel are combined to form the quality evaluation vector. ,in This represents the number of sensor channels.

[0031] The data quality assessment module also includes a spatiotemporal uncertainty propagation unit. This unit is based on the topology of the sensor network (e.g., an adjacency matrix constructed based on the physical locations of the sensors). ,in Indicates sensor With sensors The spatial distance between them is the inverse normalized value, and the temporal correlation is (e.g., using a time decay factor). , (Using a time constant), a spatiotemporal propagation model is constructed. Specifically, when a certain sensor node is detected... Data quality metrics When the value is below a preset threshold, the spatiotemporal uncertainty propagation unit estimates its impact on neighboring nodes in the following manner. Degree of impact:

[0032] in Connection weights are set based on topology (e.g., according to the inverse of the physical distance between sensors or the strength of signal correlation). For nodes and The Euclidean distance or graph distance between them. This is the feature length scale parameter (which can be set empirically, such as half the distance between the two farthest nodes in the sensor network, or obtained through training with historical data). For adjacent time steps... Furthermore, a time propagation factor is introduced:

[0033] in The time decay coefficient ( The spatiotemporal uncertainty propagation unit outputs a vector of predicted impact levels. It is used by the dynamic hierarchical early warning module.

[0034] The multi-level diagnostic engine includes a first diagnostic unit, a second diagnostic unit, and a diagnostic fusion unit.

[0035] The first diagnostic unit extracts global spatiotemporal features from multi-source time-series sensor data based on graph convolutional neural networks (GCN). Specifically, the first diagnostic unit uses multi-source time-series data... As input, combined with the sensor topology structure Spatial dependencies are extracted through multi-layer graph convolution operations, and temporal dependencies are extracted through temporal convolution or LSTM, ultimately outputting the first fault probability distribution. (For example: probability vectors for fault types such as bearing failure, gear failure, and imbalance) and their corresponding first confidence levels. Confidence level It can be calculated based on the maximum probability value output by Softmax or the uncertainty estimate based on Monte Carlo dropout.

[0036] The second diagnostic unit extracts local temporal features based on a self-attention mechanism (Transformer or self-attention LSTM). Specifically, the second diagnostic unit performs self-attention encoding on the temporal data of each sensor channel to capture short-term dependencies and anomalous patterns within each channel. Then, it fuses the features of each channel through a fully connected layer to output a second fault probability distribution. and second confidence level .

[0037] Hierarchical confidence propagation: Within the first diagnostic unit, each layer of the GCN (e.g., the first layer)... (Layer) calculates the local confidence level for each sensor node. This confidence level can be calculated based on metrics such as the magnitude of the node's feature vector and its consistency with the features of neighboring nodes. Then, the local confidence level propagates forward along the network hierarchy: ,in For the first Layer feature fusion weights (e.g., graph attention coefficients). Initial confidence level. It can be set to 1, or calculated based on quality indicators of the input data. Finally, after... After layer propagation, the average confidence score of all nodes in the last layer is taken as the corrected first confidence score. The second diagnostic unit employs a similar mechanism, propagating confidence scores between self-attention layers. This hierarchical confidence propagation ensures that the diagnostic confidence scores reflect the quality of feature fusion in the intermediate layers.

[0038] The diagnostic fusion unit will distribute the first fault probability. Second Fault Probability Distribution Perform weighted fusion to generate a comprehensive failure probability. ,in For fusion weights. Fusion weights The weight of the second diagnostic unit can be dynamically adjusted based on the quality evaluation vector output by the data quality assessment module: when the completeness or signal-to-noise ratio in the quality evaluation vector is low, the weight of the second diagnostic unit is increased (because local temporal features are relatively insensitive to data missingness); when the data quality is high, the weight of the first diagnostic unit is increased (to make full use of global structural information).

[0039] The dynamic hierarchical early warning module includes a dynamic threshold calculation unit and an early warning generation unit.

[0040] The dynamic threshold calculation unit receives the quality evaluation vector. First confidence level Second confidence level weighted value And the predicted impact of the spatiotemporal uncertainty propagation unit output. Calculate dynamic early warning threshold An exemplary calculation formula is a multiplicative model to ensure that the threshold is always within a reasonable range:

[0041] in, The basic threshold (can be set according to the failure probability statistics under normal operating conditions of the equipment). This represents the mean or weighted sum of the quality evaluation vectors. The mean of the estimated impact level, These are preset weighting coefficients. These coefficients can be obtained through training and optimization using historical data, for example, by using grid search or Bayesian optimization methods to minimize the false positive and false negative rates. When at least one component of the quality assessment vector falls below a preset lower threshold (e.g., completeness < 0.5), the dynamic threshold calculation unit automatically increases the threshold. To reduce the sensitivity of early warnings and avoid false alarms caused by low-quality data. When the first or second confidence level is lower than the preset confidence threshold (e.g., When this occurs, the dynamic threshold calculation unit sends a forced re-diagnosis command to the multi-level diagnostic engine, requesting that the diagnostic process be re-executed.

[0042] The early warning generation unit will comprehensively consider the probability of failure. With dynamic early warning threshold Comparison. When At that time, a graded early warning signal is generated. The graded early warning signal may include information such as the early warning level (e.g., prompt level, warning level, emergency level), the probability of failure prediction, the identification of the affected components, and the recommended handling time.

[0043] The closed-loop feedback unit records the actual fault confirmation result after each warning is triggered (e.g., the actual fault type reported by maintenance personnel, maintenance records, etc.) and uses it as a feedback signal. Based on the feedback result, the closed-loop feedback unit adjusts the threshold adjustment parameters in the dynamic threshold calculation unit (e.g., adjusting...). coefficient, or adjust the base threshold This forms an adaptive closed-loop adjustment of the warning threshold. For example, if the actual confirmed fault matches the warning result, but the warning threshold is too high, leading to missed reports, the threshold should be appropriately lowered. Conversely, if false alarms occur frequently, then increase... Alternatively, increase the penalty coefficient for the quality evaluation vector.

[0044] Example 2: Confidence Cross-Validation and Credibility Validation Based on the above embodiment 1, this embodiment further adds a confidence cross-validation unit and a diagnostic result confidence verification unit.

[0045] The confidence cross-validation unit is connected between the first diagnostic unit, the second diagnostic unit, and the diagnostic fusion unit. It calculates the first fault probability distribution. With the second failure probability distribution Similarity between them, for example, using KL divergence or cosine similarity:

[0046] when If the consensus value falls below a preset consensus threshold (e.g., 0.7), the current diagnostic result is deemed to have uncertainty. The confidence cross-validation unit uses this uncertainty as a penalty factor. (For example This data is input to the diagnostic fusion unit. When calculating the overall failure probability, the diagnostic fusion unit multiplies the penalty factor by the original fusion factor. Overall result: This reduces the output value and avoids giving an excessively high probability of failure when the diagnosis is inconsistent.

[0047] The diagnostic result credibility verification unit connects the diagnostic fusion unit and the dynamic hierarchical early warning module. Based on historical fault confirmation results accumulated in the closed-loop feedback unit (including actual fault confirmation results after each early warning trigger and operational data confirming no faults under normal operating conditions), it assesses the credibility of the current comprehensive diagnostic result. Specifically, the credibility verification unit statistically analyzes the fault type distribution in historical fault confirmation results and calculates the current diagnostic result. The deviation between the corresponding fault type and the historical distribution (e.g., using JS divergence). When the deviation exceeds a preset deviation threshold (e.g., 0.3), the credibility verification unit reduces the credibility correction coefficient. Output values ​​(e.g.) When the credibility correction coefficient When the confidence level falls below a preset threshold (e.g., 0.7), a re-diagnosis trigger command is sent to the multi-level diagnostic engine, and... As a penalty factor, it is input into the diagnostic fusion unit to dynamically adjust the fusion weights of subsequent diagnostic results (e.g., update). ).

[0048] Example 3: Progressive Diagnosis and Adaptive Fusion This embodiment further optimizes the computational efficiency of the multi-level diagnostic engine.

[0049] Progressive diagnostic mode: The multi-level diagnostic engine is configured to support progressive diagnostics. This is achieved when the dynamic hierarchical early warning module determines the overall fault probability. Exceeding the preset first warning threshold (For example However, it did not exceed the dynamic early warning threshold. (For example When a fault occurs, instead of directly triggering a full diagnosis, an intermediate-level diagnosis is triggered first. The computational complexity of the intermediate-level diagnosis is lower than that of the full diagnosis; for example, it may involve running only a lightweight version of the first or second diagnostic unit (such as a single-layer GCN or simplified self-attention), or analyzing only data from some key sensor channels. If the intermediate-level diagnosis outputs an intermediate fault probability... Still exceeds the first warning threshold Then, a complete diagnostic is triggered (i.e., running two diagnostic units simultaneously and fusing them). This progressive strategy can effectively reduce the computational overhead of the system under most normal operating conditions.

[0050] Differential-driven adaptive weight adjustment: The diagnostic fusion unit is also configured to monitor the probability distribution of the first fault. With the second failure probability distribution The difference between the two patterns is assessed (e.g., using Euclidean distance or KL divergence). When the difference exceeds a preset difference threshold (e.g., 0.4), an adaptive weight adjustment mechanism is activated. This mechanism dynamically optimizes the fusion weights by minimizing the deviation between the weighted fusion result and the matching patterns in the historical fault mode library. The historical fault mode library stores fault samples confirmed in past diagnoses and their corresponding optimal fusion weights. Adaptive adjustment can employ gradient descent or Bayesian optimization methods, with the objective function being:

[0051] in This represents the probability distribution of the fault mode in the historical fault mode library that best matches the current input. After optimization, a new... The value is used for subsequent fusion calculations.

[0052] Example 4: Model Optimization and Data Management This embodiment further describes the model optimization and data management mechanisms.

[0053] Model Optimization Subunit: The closed-loop feedback unit also includes a model optimization subunit. This subunit uses the complete fault cycle data collected after each warning trigger (e.g., multi-source sensor data for 10 minutes before and after the warning) as training samples, and periodically (e.g., weekly) incrementally updates the diagnostic models of the first and second diagnostic units. Incremental updates can employ online learning algorithms (such as Elastic Weight Consolidation (EWC) or incremental gradient descent) to avoid catastrophic forgetting. The updated model parameters are fed back to the dynamic threshold calculation unit to adjust the base threshold for the dynamic warning threshold. (For example, recalibrating based on the average output probability of the new model).

[0054] Abnormal data capture and upload unit: The multi-level diagnostic engine also includes an abnormal data capture and upload unit. When considering the overall failure probability... When the probability of any fault type in the fault probability distribution output by the first or second diagnostic unit exceeds a preset threshold (e.g., 0.5), the abnormal data capture and upload unit captures sensor data segments of a set time length (e.g., 5 minutes before and after) before and after the trigger time, and uploads them as high-value sample data to the cloud or local storage unit. This sample data can be used for subsequent model optimization and updates, expansion of the fault mode library, and algorithm verification. Data can be anonymized and compressed before uploading to save bandwidth and storage space.

[0055] Example 5: Parameter Setting and Experimental Verification The following provides the parameter settings and experimental verification results for a specific example (exemplary data; in actual applications, these can be adjusted according to the characteristics of the device).

[0056] Sensor configuration: Install 3 acceleration vibration sensors (sampling frequency 10kHz) in the motor bearing housing, 2 temperature sensors (sampling frequency 1Hz) in the motor stator, and 1 current sensor (sampling frequency 1kHz) in the power line.

[0057] Diagnostic unit configuration: The first diagnostic unit adopts a 2-layer graph convolutional network with a hidden layer dimension of 64 and a time window length of 128; the second diagnostic unit adopts a 4-head self-attention mechanism with a encoder layer of 2 and a feedforward network dimension of 128.

[0058] Dynamic threshold parameter: Multiplicative model parameter set to The upper limit is truncated to 0.95. Preset lower thresholds: data integrity lower limit 0.6, signal-to-noise ratio lower limit 10dB. Preset reliability lower limit: or A forced re-diagnosis is triggered when the value is below 0.6.

[0059] Experimental Dataset Description: This experiment uses a simulated motor bearing fault dataset, which includes four states: normal, inner race fault, outer race fault, and rolling element fault. 100 samples are collected for each state, and each sample lasts for 10 seconds.

[0060] Simulation Comparison: On a simulated motor bearing fault dataset, with 30% of sensor data randomly missing, the system of this invention was compared with an existing equipment monitoring and prediction scheme employing a three-layer prediction architecture (edge ​​layer threshold rule model, middle layer XGBoost model, and cloud layer temporal convolutional network model). Simulation results show that the fault diagnosis accuracy of the system of this invention reaches 94.2%, while the comparison scheme is only 78.6%; in terms of false alarm rate, the system of this invention has a false alarm rate of 3.5%, while the comparison scheme has a false alarm rate of 12.3%. In a scenario where all sensor nodes fail, the spatiotemporal uncertainty propagation unit of this invention can automatically increase the dynamic early warning threshold by approximately 20%, reducing the false alarm rate from 12.1% to 4.2%, verifying the robustness of this invention.

[0061] The above detailed embodiments describe multiple examples of the present invention, including system architecture, core algorithms, parameter settings, and experimental verification. Those skilled in the art should understand that these embodiments are for illustrative purposes only and are not intended to limit the scope of protection of the present invention. Any simple modifications, equivalent variations, and alterations made to the above embodiments based on the technical essence of the present invention fall within the scope of protection of the present invention.

Claims

1. A smart fault diagnosis and early warning system for electromechanical equipment based on multi-dimensional sensing, characterized in that, include: A multi-dimensional sensing data acquisition module is used to acquire multi-source time-series sensing data from multiple heterogeneous sensors on electromechanical equipment; The data quality assessment module, connected to the multi-dimensional sensor data acquisition module, is used to perform quality analysis on the multi-source time-series sensor data and generate a quality evaluation vector. The data quality assessment module also includes a spatiotemporal uncertainty propagation unit, which is used to estimate the impact of data quality degradation on the diagnostic confidence of adjacent sensor nodes and adjacent time steps based on the topology and temporal correlation of the sensor network, and output the impact prediction result. A multi-level diagnostic engine, connected to the data quality assessment module, includes: The first diagnostic unit is used to output a first fault probability distribution and a first confidence level based on multi-source time-series sensor data. The second diagnostic unit is used to output a second fault probability distribution and a second confidence level based on multi-source time-series sensor data. The diagnostic fusion unit is used to fuse the first fault probability distribution and the second fault probability distribution to generate a comprehensive fault probability. The first diagnostic unit and the second diagnostic unit are each internally configured to perform hierarchical confidence propagation and output the first confidence level and the second confidence level after hierarchical confidence propagation correction. A dynamic hierarchical early warning module, connected to the data quality assessment module and the multi-level diagnostic engine, includes: The dynamic threshold calculation unit is used to calculate the dynamic early warning threshold based on the quality evaluation vector, the weighted value of the first confidence level and the second confidence level, and the influence degree prediction result. The early warning generation unit is used to generate a graded early warning signal when the overall failure probability exceeds the dynamic early warning threshold. The closed-loop feedback unit is used to correct the threshold adjustment parameters of the dynamic threshold calculation unit based on the actual fault confirmation results after the warning is triggered, so as to form an adaptive closed-loop adjustment of the warning threshold.

2. The intelligent fault diagnosis and early warning system for electromechanical equipment based on multi-dimensional sensing according to claim 1, characterized in that, The quality evaluation vector includes at least a data integrity index and a signal-to-noise ratio index.

3. The intelligent fault diagnosis and early warning system for electromechanical equipment based on multi-dimensional sensing according to claim 1, characterized in that, The first diagnostic unit extracts the global structural features of the multi-source time-series sensor data based on a first type of feature extraction algorithm to output the first fault probability distribution; the second diagnostic unit extracts the local time-series features of the multi-source time-series sensor data based on a second type of feature extraction algorithm to output the second fault probability distribution.

4. The intelligent fault diagnosis and early warning system for electromechanical equipment based on multi-dimensional sensing according to claim 1, characterized in that, The dynamic threshold calculation unit is further configured to: automatically increase the dynamic warning threshold to reduce warning sensitivity when at least one component in the quality evaluation vector is lower than a preset lower threshold; and send a forced re-diagnosis command to the multi-level diagnostic engine when the first confidence level or the second confidence level is lower than a preset confidence lower limit.

5. The intelligent fault diagnosis and early warning system for electromechanical equipment based on multi-dimensional sensing according to claim 1, characterized in that, The multi-level diagnostic engine also includes a confidence cross-validation unit connected between the first diagnostic unit, the second diagnostic unit, and the diagnostic fusion unit. It is used to calculate the similarity between the first fault probability distribution and the second fault probability distribution. When the similarity is lower than a preset consistency threshold, it is determined that the current diagnostic result has uncertainty. The uncertainty identifier is used as a penalty factor when the diagnostic fusion unit performs weighted fusion to reduce the output value of the comprehensive fault probability.

6. The intelligent fault diagnosis and early warning system for electromechanical equipment based on multi-dimensional sensing according to claim 1, characterized in that, The multi-level diagnostic engine also includes a diagnostic result credibility verification unit, connected between the diagnostic fusion unit and the dynamic hierarchical early warning module. This unit evaluates the credibility of the current comprehensive diagnostic result output by the diagnostic fusion unit based on historical fault confirmation results accumulated in the closed-loop feedback unit, and outputs a credibility correction coefficient. The historical fault confirmation results include actual fault confirmation results after each early warning trigger and operational data confirming no faults under normal operating conditions. The credibility verification unit is configured to reduce the output value of the credibility correction coefficient when the deviation between the fault type in the current comprehensive diagnostic result and the fault type distribution in the historical fault confirmation results exceeds a preset deviation threshold. When the confidence correction coefficient is lower than the preset confidence lower limit threshold, a re-diagnosis trigger instruction is sent to the multi-level diagnostic engine, and the confidence correction coefficient is input as a penalty factor into the diagnostic fusion unit for dynamic correction of the fusion weight of subsequent diagnostic results.

7. The intelligent fault diagnosis and early warning system for electromechanical equipment based on multi-dimensional sensing according to claim 1, characterized in that, The multi-level diagnostic engine is configured to support a progressive diagnostic mode: when the dynamic hierarchical early warning module determines that the overall fault probability exceeds a preset first early warning threshold but does not exceed the dynamic early warning threshold, the intermediate-level diagnosis of the multi-level diagnostic engine is triggered. The computational complexity of the intermediate-level diagnosis is lower than that of the complete diagnosis but higher than that of the independent diagnosis of the first or second diagnostic unit. When the intermediate fault probability output by the intermediate-level diagnosis still exceeds the first early warning threshold, the complete diagnosis is triggered again.

8. The intelligent fault diagnosis and early warning system for electromechanical equipment based on multi-dimensional sensing according to claim 1, characterized in that, The diagnostic fusion unit is configured to: when the difference between the first fault probability distribution and the second fault probability distribution exceeds a preset difference threshold, activate a weight adaptive adjustment mechanism to dynamically optimize the fusion weights of the first diagnostic unit and the second diagnostic unit by minimizing the deviation between the weighted fusion result and the matching pattern in the historical fault mode library.

9. The intelligent fault diagnosis and early warning system for electromechanical equipment based on multi-dimensional sensing according to claim 1, characterized in that, The closed-loop feedback unit further includes a model optimization subunit, which uses the complete fault cycle data collected after each warning is triggered as training samples to periodically update the diagnostic models of the first diagnostic unit and the second diagnostic unit, and feeds back the updated model parameters to the dynamic threshold calculation unit to adjust the base threshold of the dynamic warning threshold.

10. The intelligent fault diagnosis and early warning system for electromechanical equipment based on multi-dimensional sensing according to claim 1, characterized in that, The multi-level diagnostic engine also includes an abnormal data interception and uploading unit, which is used to intercept sensor data segments of a set time length before and after the triggering time when the overall fault probability exceeds a preset threshold, or when the probability of any fault type in the fault probability distribution output by the first diagnostic unit or the second diagnostic unit exceeds a preset fault identification threshold, and upload them as high-value sample data to the cloud or storage unit for subsequent optimization and updates of the diagnostic model.

Images