Tunnel electromechanical equipment bimodal fault diagnosis method

Abstract

The invention discloses a dual-mode fault diagnosis method for tunnel electromechanical equipment, and relates to the technical field of tunnel electromechanical equipment. The technical problems that a single diagnosis model is difficult to give consideration to emergency fault real-time response and progressive fault accurate identification, a static threshold is weak in adaptive tunnel complex environment capability, misinformation and missing detection are easily caused, single signal analysis cannot effectively identify composite faults, and a maintenance mode is lagged due to lack of early warning are solved. Emergency faults such as overload and overheating are quickly captured through the first diagnosis model, historical data are deeply analyzed through the second diagnosis model based on machine learning, hidden progressive faults such as bearing abrasion and blade cracks are accurately recognized, the defect of a single model is overcome, the fault recognition accuracy is greatly improved, the false alarm rate is reduced, meanwhile, a static threshold value is abandoned, and the fault recognition efficiency is improved. The threshold value and the diagnosis weight are dynamically adjusted according to the real-time operation state of the equipment to adapt to complex environments such as tunnel temperature and humidity fluctuation and load abrupt change, and normal fluctuation misjudgment is avoided.

Description

Technical Field

[0001] This invention relates to the field of tunnel electromechanical equipment technology, specifically a dual-mode fault diagnosis method for tunnel electromechanical equipment. Background Technology

[0002] With the rapid development of modern transportation infrastructure, tunnel electromechanical equipment (such as ventilation fans, lighting fixtures, and power systems) plays a crucial role in ensuring the safe operation of tunnels. However, traditional fault diagnosis methods often rely on manual experience or a single model, which has the following significant drawbacks: Existing technologies largely rely on human experience or single fault diagnosis models, which have significant drawbacks: First, single models have limitations. While rule-based models can quickly respond to emergency faults, they cannot identify progressive faults that require long-term trend analysis. Although machine learning-based models can capture complex fault patterns, their real-time response latency is high, making it difficult to meet the real-time judgment requirements of emergency faults. Second, traditional static threshold configurations cannot adapt to the complex environment of tunnels and the dynamic operating status of equipment, easily leading to misjudgments of normal fluctuations or missed detections of minor equipment degradation, thus creating potential safety hazards. Third, the isolated feature analysis mode only extracts single-dimensional signal features, lacking comprehensive mining and fusion of multi-domain features. It is difficult to distinguish between instantaneous fluctuations and long-term degradation, and it is also unable to accurately capture the collaborative features of complex faults, resulting in low diagnostic accuracy. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention provides a dual-mode fault diagnosis method for tunnel electromechanical equipment. This method solves the problems of single diagnostic models being unable to simultaneously address real-time response to emergency faults and accurate identification of progressive faults, weak adaptability of static thresholds to the complex tunnel environment leading to false alarms and missed detections, inability of single signal analysis to effectively identify complex faults, and lack of early warning resulting in delayed maintenance modes.

[0004] To achieve the above objectives, the present invention provides a dual-mode fault diagnosis method for tunnel electromechanical equipment, comprising the following steps: (a) The real-time acquired tunnel motor equipment operation data is preprocessed and feature extracted to analyze its short-term dynamics. Time-domain / instantaneous correlation features are extracted to generate real-time fault feature vectors corresponding to the instantaneous state of the equipment. Frequency domain transformation and sliding window method are applied to the operation data to extract historical sequence features and generate trend fault feature vectors. The operation data includes vibration signals, temperature signals and current signals. (b) Input the real-time fault vector into the first diagnostic model, obtain the corresponding fault judgment through threshold comparison, and generate the first fault diagnosis result; (c) Input the trend fault feature vector into the second diagnostic model, identify its hidden feature faults based on the feature mapping relationship trained in history, and generate the second fault diagnosis result; (d) Based on the first fault diagnosis result and the second fault diagnosis result, a comprehensive judgment is made according to the preset fusion rules to obtain the corresponding fault type and generate a comprehensive fault diagnosis report; the fusion rules include: if the first fault diagnosis result indicates an emergency fault, the result is directly adopted as the comprehensive judgment result; otherwise, the first fault diagnosis result and the second fault diagnosis result are weighted and fused to generate the comprehensive fault diagnosis report. (e) Based on the generated comprehensive fault diagnosis report, trigger the corresponding alarm and match the fault knowledge base to output the corresponding maintenance suggestions.

[0005] As a further aspect of the present invention: the preprocessing and feature extraction in step (a) include: The vibration signal is filtered and denoised, and the temperature and current signals are standardized to eliminate the dimensional differences of environmental noise. The peak factor Cf and waveform factor Wf of the vibration signal, the real-time effective value Ir and fluctuation energy ΔIr of the current signal, and the instantaneous rate of change ΔTr of the temperature signal are extracted and combined to form a short-term dynamic feature set Sf. The short-term dynamic feature set Sf is then integrated in sequence to generate a real-time fault feature vector Vr. The vibration signal is frequency-domain transformed to extract the low-frequency energy ratio Ef, the temperature signal is analyzed by sliding window to extract the mean Tavg and the baseline offset ΔTbase, and the harmonic distortion rate mean THDavg is extracted from the current signal. These are combined to form a long-term trend feature set Lf. The long-term trend feature set Lf is then integrated in sequence to generate a trend fault feature vector Vt.

[0006] As a further aspect of the present invention: the rule-based first diagnostic model in step (b) specifically performs the following determination: Extract the effective value of current Ir from the real-time fault feature vector Vt, and compare it with the safe current threshold dynamically configured according to the equipment type and operating conditions. If the current exceeds the limit for a first preset time, it is determined to be an overload fault. The vibration peak factor Cf in the real-time fault feature vector Vt is extracted and compared with a preset dynamic threshold. If it continues to exceed the limit for a second preset duration, it is determined to be a mechanical loosening fault. The temperature value in the real-time fault feature vector Vt is extracted and compared with the preset safe temperature range. If the temperature exceeds the limit for a third preset duration, it is determined to be an overheating fault.

[0007] As a further aspect of the present invention: the second diagnostic model in step (c) is a convolutional neural network (CNN) model; The CNN model uses vibration spectrum, temperature time series and current time series extracted from equipment operation data as training samples, and uses the corresponding fault category as a label for training. The model learns the mapping relationship between the trend fault feature vector and the potential fault type to predict the potential fault type, which includes bearing wear, blade cracks and electrical faults.

[0008] As a further aspect of the present invention: in step (d), when the first fault diagnosis result does not indicate an emergency fault, the specific method of the comprehensive weighted calculation is as follows: Based on the current load change rate of the equipment, the first weight W1 of the first fault diagnosis result and the second weight W2 of the second fault diagnosis result are dynamically allocated, wherein when the load change rate is greater than a set threshold, the proportion of W1 is increased. At the same time, the corrected weight W2' is obtained by comparing the historical diagnostic accuracy of the second diagnostic model with W2. The first fault diagnosis result and the second diagnosis result are weighted and fused based on W1 and W2' to generate the comprehensive fault diagnosis result.

[0009] As a further aspect of the present invention: the comprehensive fault diagnosis report generated in step (e) includes the fault type, the severity level based on confidence level, and the fault evolution trend predicted based on long-term trend characteristics.

[0010] As a further aspect of the present invention: the tunnel electromechanical equipment includes a ventilation fan and lighting fixtures; The sensor deployment location in step (a), the dynamic threshold in step (c), and the training samples of the second diagnostic model in step (d) are all configured differently for the physical structure, operating characteristics, and typical fault modes of ventilation fans or lighting fixtures.

[0011] As a further aspect of the present invention: in step (a), vibration signals are collected by an acceleration sensor installed in the bearing of the equipment, temperature signals are collected by a temperature sensor embedded in the stator of the motor, and current signals are collected by a non-contact current sensor based on the Hall effect. The collected raw data is transmitted in real time to edge computing devices or cloud servers for processing via wireless communication modules or wired industrial networks.

[0012] As a further aspect of the present invention: the alarm triggering in step (e) specifically refers to: If the comprehensive fault diagnosis report indicates an emergency fault, an on-site audible and visual alarm will be triggered, and a warning message will be sent to the remote monitoring center and the mobile terminals of maintenance personnel at the same time. If the indication is a potential fault, only send an early warning message to the remote monitoring center; The maintenance recommendations are generated by querying the equipment maintenance knowledge base, which stores the mapping relationship between historical fault types and verified maintenance measures.

[0013] A dual-mode fault diagnosis system for tunnel electromechanical equipment, the system comprising: The data acquisition and transmission module is used to collect vibration, temperature and current signals through sensors deployed on the equipment and transmit the data to the computing unit in real time; The feature extraction and construction module is used to preprocess the received data and extract short-term dynamic features and long-term trend features respectively to construct real-time fault feature vectors and trend fault feature vectors. The real-time rule-based diagnosis module has a built-in first diagnosis model based on dynamic threshold rules, which is used to receive the real-time fault feature vector and output the first fault diagnosis result. The deep learning analysis module has a built-in second diagnostic model based on a convolutional neural network, which is used to receive the trend fault feature vector and output the second fault diagnosis result. The dynamic fusion decision module is used to comprehensively analyze the first and second fault diagnosis results according to preset fusion rules, and generate a comprehensive diagnosis report that includes fault type, severity and evolution trend.

[0014] This invention provides a dual-mode fault diagnosis method for tunnel electromechanical equipment. Compared with the prior art, it has the following advantages: (1) This invention uses a first diagnostic model to quickly capture emergency faults such as overload and overheating, and relies on a second diagnostic model based on machine learning to deeply analyze historical data, accurately identify hidden progressive faults such as bearing wear and blade cracks, make up for the shortcomings of a single model, greatly improve the accuracy of fault identification and reduce the false alarm rate. At the same time, it abandons static thresholds and dynamically adjusts the thresholds and diagnostic weights according to the real-time operating status of the equipment, adapts to complex environments such as temperature and humidity fluctuations and sudden load changes in tunnels, and avoids misjudgment of normal fluctuations. (2) This invention extracts peak factor and spectral energy distribution from vibration signals, analyzes trends from temperature signals using the sliding window method, and calculates harmonic distortion rate from current signals to achieve multi-dimensional differentiation of mechanical, electrical, and thermal faults. Through end-to-end training of multiple types of signals using CNN, it automatically mines the mapping relationship between fault features and types, adapts to nonlinear and multivariate fault modes, and significantly improves the ability to identify complex faults. (3) Based on the model, the present invention predicts the trend of changes in parameters such as temperature and vibration, and can trigger an early warning in the early stage of the fault, realizing the transformation from post-maintenance to preventive maintenance. The fault type, severity and evolution trend are displayed through the Web or mobile terminal visualization interface. Combined with historical maintenance records, targeted suggestions are generated, which can effectively shorten the fault response time, reduce operation and maintenance costs, and ensure the continuous and stable operation of tunnel electromechanical equipment. Attached Figure Description

[0015] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a system flowchart of the present invention. Detailed Implementation

[0016] 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.

[0017] Example 1 Please see Figure 1 This application provides a dual-mode fault diagnosis method for tunnel electromechanical equipment. (a) The real-time collected operating data of the tunnel motor equipment is preprocessed and features are extracted. Its short-term dynamics are analyzed, and time-domain / instantaneous correlation features are extracted to generate real-time fault feature vectors corresponding to the instantaneous state of the equipment. The operating data is frequency-domain transformed and the sliding window method is used to extract historical sequence features to generate trend fault feature vectors. The operating data includes vibration signals, temperature signals and current signals. (b) Input the real-time fault vector into the first diagnostic model, obtain the corresponding fault judgment through threshold comparison, and generate the first fault diagnosis result; (c) Input the trend fault feature vector into the second diagnostic model, identify its hidden feature faults based on the feature mapping relationship trained in history, and generate the second fault diagnosis result; (d) Based on the first fault diagnosis result and the second fault diagnosis result, a comprehensive judgment is made according to the preset fusion rules to obtain the corresponding fault type and generate a comprehensive fault diagnosis report; the fusion rules include: if the first fault diagnosis result indicates an emergency fault, the result is directly adopted as the comprehensive judgment result; otherwise, the first fault diagnosis result and the second fault diagnosis result are weighted and fused to generate the comprehensive fault diagnosis report. (e) Based on the generated comprehensive fault diagnosis report, trigger the corresponding alarm and match it with the fault knowledge base to output the corresponding maintenance suggestions: Example 2 Furthermore, based on Embodiment 1, a dual-mode fault diagnosis method for tunnel electromechanical equipment is described in more detail, specifically including the following steps: Step 1: Comprehensively collect the core status signals of the tunnel's electromechanical equipment during operation to provide complete and real-time data support for subsequent feature extraction and fault analysis. The specific collection process is as follows: Confirm the key monitoring parameters of tunnel electromechanical systems, including the vibration signal Vt reflecting the mechanical operating status, the temperature signal Tt characterizing the thermal state of the equipment, and the current signal It reflecting the electrical load status; Signal acquisition is achieved by deploying dedicated sensors at key locations on the equipment body: acceleration sensors are installed in vibration-sensitive parts such as bearings, temperature sensors are embedded in core heat-generating components such as motor stators, and non-contact current sensors based on the Hall effect principle are used to collect current data. The raw data collected by the sensors is transmitted in real time to the edge computing device or cloud server via 4G / 5G wireless communication module or industrial Ethernet wired transmission, ensuring that the data transmission latency is controlled within 50ms and guaranteeing the real-time nature of the diagnosis. The data collection covers various types of tunnel electromechanical equipment, such as ventilation fans and lighting fixtures. For each type of equipment, the sensor deployment location and collection frequency are optimized based on its installation scenario and operating characteristics. Specifically, through the collaborative deployment of multiple types of sensors, the system comprehensively covers the three core operating dimensions of the equipment: mechanical, electrical, and thermal. This avoids missing fault information due to single signal acquisition, laying a data foundation for subsequent accurate diagnosis. Simultaneously, a highly reliable communication transmission scheme is used to ensure data real-time performance and integrity, meeting the fault diagnosis needs in the complex environment of tunnels.

[0018] Step 2: Denoise and normalize the collected raw data, and extract short-term dynamic features and long-term trend features through multi-domain analysis to construct real-time fault feature vectors and trend fault feature vectors, respectively. The specific implementation method is as follows: Data preprocessing: The vibration signal Vt is denoised using a Butterworth filter to remove noise caused by environmental dust and airflow interference. The temperature signal Tt and current signal It are normalized using the Z-score normalization method to eliminate dimensional differences under different equipment and operating conditions. Short-term dynamic feature extraction: For the preprocessed vibration signal, calculate the peak factor Cf, waveform factor Wf and other time-domain features; for the current signal, calculate the real-time effective value Ir, instantaneous fluctuation ΔIr and other features; for the temperature, calculate the instantaneous rate of change ΔTr and other features. The above features are labeled as the short-term dynamic feature set Sf={Cf, Wf, Ir, ΔIr, ΔTr}. Long-term trend feature extraction: The vibration signal is transformed into the frequency domain through the Fast Fourier Transform (FFT), and the proportion of energy Ef in the low-frequency band (<100Hz) is extracted. For the temperature signal, the sliding window method (window size is 10 minutes) is used to calculate the mean value Tavg and the reference offset △Tbase (△Tbase = |Tavg - T0|, where T0 is the normal operating temperature baseline of the device). For the current signal, the average value of the harmonic distortion rate THD within 1 hour, THDavg, is calculated. The above features are calibrated as the long-term trend feature set Lf = {Ef, Tavg, △Tbase, THDavg}; Feature vector construction: The short-term dynamic feature set Sf is integrated in order to generate the real-time fault feature vector Vr[Cf, Wf, Ir, △Ir, △Tr]. The long-term trend feature set Lf is integrated in order to generate the trend fault feature vector Vt = [Ef, Tavg, △Tbase, THDavg]. Specifically, interference and dimensional differences are eliminated through standardized preprocessing, and multi-domain feature extraction is used to accurately characterize instantaneous faults (such as sudden overload) and progressive faults (such as bearing wear). The dual feature vectors provide targeted inputs for the subsequent dual-modal model, ensuring that different types of fault features are not missed.

[0019] Step three: Input the real-time fault feature vector Vr into the first diagnostic model based on dynamic rules. Through threshold comparison and duration determination, emergency faults such as overload, mechanical looseness, and overheating are quickly identified. Specifically: Dynamic threshold configuration: Based on the device type, preset the basic thresholds. The safety threshold of the ventilation fan current Ith = 1.2 - 1.3 times the rated current (dynamically adjusted when the load changes, and the upper limit is increased to 1.3 times under high-load conditions), the vibration peak factor threshold Cth = 3.5, and the temperature safety range [Tmin, Tmax] (for the ventilation fan bearing, Tmax = 80°C; for the lighting fixture, Tmax = ambient temperature + 40°C); Overload fault determination: Extract the effective current value Ir in the real-time fault feature vector Vr and compare it with the preset safety threshold Ith. If Ir > Ith and the duration ≥ 10 minutes, it is determined as an overload fault, and the fault representation F1 = 1 is generated. If Ir ≤ Ith or the duration < T1, then F1 = 0; Mechanical looseness fault determination: Extract the vibration peak factor Cf in Vr and compare it with the dynamic threshold Cth. If Cf > Cth and the duration T2 ≥ 5 minutes, it is determined as a potential mechanical looseness fault, and the fault identifier F2 = 1 is generated. If Cf ≤ Cth or the duration < T2, then F2 = 0; Overheating fault determination: Extract the temperature signal Tr in Vr and compare it with the preset range [Tmin, Tmax]. If Tr > Tmax or Tr << Tmin and the duration T3 ≥ 3 minutes, it is determined as an overheating fault, and a fault flag F3 = 1 is generated. If Tr is within [Tmin, Tmax] or the duration < T3, then F3 = 0; Output of the first diagnostic result: Integrate the fault flags to generate the first diagnostic result R1, where R1 = 1 (overload fault), R1 = 2 (mechanical looseness fault), R1 = 3 (overheating fault), R1 = 0 (no emergency fault), and key information such as the fault duration and over-limit amplitude are recorded synchronously.

[0020] Step 4: Input the trend fault feature vector Vt into the second diagnostic model based on the convolutional neural network (CNN). Through the feature mapping relationship trained by historical data, identify hidden and progressive faults such as bearing wear, blade rupture, and electrical faults. The specific implementation is as follows: Model construction: The CNN model includes an input layer, 2 convolutional layers, 2 pooling layers, 1 fully connected layer, and an output layer. The input data of the input layer is a three-dimensional tensor (number of channels 4 × time steps 60 × number of features 4, corresponding to 4 long-term trend features and 60 time node data); Based on the historical operation data of the equipment, extract the vibration spectrogram Vspec (128 × 128 pixels), temperature sequence Tseq (60 × 1 dimension), and current sequence lseq (60 × 1 dimension) as training samples, and label the fault label L (where L = 1 represents bearing wear, L2 = 2 represents blade crack, L3 = 3 represents electrical fault, L0 = 0 represents no fault); Model training: Expand the sample set through data augmentation techniques such as noise injection (adding ±5% random noise) and time offset (±3 time steps), and train the model using the Adam and cross-entropy loss functions. Iterate 100 rounds until the loss function converges (loss value < 0.05) to enable the model to automatically learn the mapping relationship between trend features and fault labels; Fault prediction: Input the trend fault feature vector Vt into the trained CNN model. The model outputs the fault prediction confidence P (P ∈ [0, 1]) and the predicted label Lpred (such as no potential fault, bearing wear fault, blade crack fault, electrical fault, etc.). Set the confidence threshold Pth = 0.8. If P ≥ Pth, then the second diagnostic result R2 = Lpred. If P < Pth, then R2 = 0 (temporarily do not determine potential faults).

[0021] [[ID=???]] Step 5: According to the preset fusion rule, dynamically adjust the weights in combination with the equipment operation status, and fuse the first diagnostic result R1 and the second diagnostic result R2 to generate a comprehensive and reliable comprehensive diagnostic report. The specific implementation is as follows: Fusion rule setting: If R1∈{1,2,3} (an urgent fault exists), then R1 is directly adopted as the comprehensive diagnostic result R without weighted calculation; if R1=0 (no urgent fault exists), then the weighted fusion rule is used to calculate R. Dynamic weight adjustment: Set the weight W1 for the first diagnostic result and the weight W2 for the second diagnostic result (W1+W2=1). W1 is quantified based on the equipment's operational stability S (S=△Load / △t, where △Load is the load change and △t is the change duration). When S>10% (load sudden change / acceleration phase), W1=0.7 and W2=0.3. When S≤10% (stable operation phase), W1=0.5 and W2=0.5. W2 is further corrected by combining the fault history confidence C (C=number of correct diagnoses in history / total number of diagnoses). Finally, W2'=W2×C. Weighted fusion calculation: The comprehensive diagnostic result R = W1 × R1 + W2' × R2. If R ≥ 1 and is an integer, the corresponding fault type is determined. If R is not an integer, the closest integer is taken as the fault type. The comprehensive confidence level Pc = W1 × P1 + W2' × P is calculated simultaneously (where P1 is the confidence level of the first diagnostic result, P1 = 1 for emergency faults and P1 = 0.9 for no emergency faults). Secondary inspection of equipment type: For ventilation fans, if R=2 (mechanical loosening), the judgment needs to be corrected based on the impeller speed (when the speed is >1500rpm, the confidence level Pc increases by 10%). For lighting fixtures, if R=3 (electrical fault), the current harmonic distortion rate (THD) needs to be further verified (when THD>10%, the fault is confirmed). The final output is a comprehensive report based on fault type (emergency fault / potential fault), severity (red level: emergency fault; yellow level: potential fault with confidence of 0.8-0.9; blue level: potential fault with confidence of 0.6-0.8), and fault evolution trend (predicting fault development in the next 24 hours based on Lf features).

[0022] Step Six: When the comprehensive diagnostic result R≠0, activate the tiered alarm mechanism and generate targeted maintenance suggestions based on the equipment maintenance knowledge base to assist maintenance personnel in quick response. The specific implementation method is as follows: Tiered alarm triggering: If a red fault (emergency fault) is triggered, the on-site audible and visual alarm will be activated to issue an alarm, and a real-time warning (including faulty device ID, type, location, and over-limit data) will be pushed to the remote monitoring center via the MQTT protocol. At the same time, an SMS will be sent to the corresponding maintenance manager. If a yellow / blue fault (potential fault) is triggered, only the warning information will be pushed to the remote monitoring center, and the on-site audible and visual alarm will not be activated. Maintenance Knowledge Base Access: Construct an equipment maintenance knowledge base KB to store historical fault records and corresponding effective maintenance measures. The format is KB={(F1, M1), (F1, M1), ...} (e.g., F1=overload fault and corresponding M1=reduce equipment load, check for short circuit in power supply line, F2=bearing wear and corresponding M2=replace bearing of the same model and apply special grease). Based on the comprehensive diagnostic result R, match the fault type F in the knowledge base KB, extract the corresponding maintenance measures M, and optimize the suggestions by combining the severity of the fault and the evolution trend (e.g., for Huangji bearing wear fault, it is recommended to arrange a shutdown to replace the bearing within 3 days; for Hongji overheat fault, it is recommended to shut down immediately, clean the heat dissipation channel and check the cooling system). Visualization and Data Storage: Through a web / mobile visualization interface, the system displays fault types, severity, evolution trends (line graphs showing changes in features such as Tavg and Ef), maintenance suggestions, and knowledge base matching records. The system also synchronously stores comprehensive diagnostic reports, raw data, model output results, and maintenance records into the database for subsequent model optimization and fault rule analysis.

[0023] Specifically, during use, vibration signals, temperature signals, and current signals are first collected in real time by accelerometers, temperature sensors, and current sensors deployed on the device itself. The data is then transmitted to edge computing devices or cloud servers via 4G / 5G communication modules. The computing unit performs noise reduction and normalization on the raw data, and extracts features by combining time domain analysis, frequency domain analysis, and time series trend analysis to generate real-time fault feature vectors containing short-term dynamic features and trend fault feature vectors containing long-term trend features. Subsequently, the real-time fault feature vectors are input into a rule-based first diagnostic model, which outputs the first fault diagnosis result through dynamic threshold comparison. The trend fault feature vector is then input into the second diagnostic model based on a convolutional neural network. The fault classifier trained on historical data identifies potential problems such as bearing wear, blade cracks, or electrical faults. The data fusion and decision module makes a comprehensive judgment based on preset rules: if the first diagnostic model detects an emergency fault, it directly triggers the on-site audible and visual alarm and sends a remote warning. Otherwise, by dynamically adjusting the weights of the two diagnostic results, a comprehensive diagnostic report is generated that includes the fault type, severity, and evolution trend. Finally, the system displays the diagnostic results through a visual interface and generates targeted suggestions based on historical maintenance measures retrieved from the knowledge base, enabling rapid fault response and precise maintenance.

[0024] Example 3 Furthermore, refer to Figure 2 As shown, a dual-mode fault diagnosis system for tunnel electromechanical equipment is proposed to implement the dual-mode fault diagnosis method for tunnel electromechanical equipment as described above, including: The data acquisition and transmission module is used to collect vibration, temperature and current signals through sensors deployed on the equipment and transmit the data to the computing unit in real time; The feature extraction and construction module is used to preprocess the received data and extract short-term dynamic features and long-term trend features respectively to construct real-time fault feature vectors and trend fault feature vectors. The real-time rule-based diagnosis module has a built-in first diagnosis model based on dynamic threshold rules, which is used to receive the real-time fault feature vector and output the first fault diagnosis result. The deep learning analysis module has a built-in second diagnostic model based on a convolutional neural network, which is used to receive the trend fault feature vector and output the second fault diagnosis result. The dynamic fusion decision module is used to comprehensively analyze the first and second fault diagnosis results according to preset fusion rules, and generate a comprehensive diagnosis report that includes fault type, severity and evolution trend.

[0025] Some of the data in the above formulas are numerical calculations with dimensions removed, and the contents not described in detail in this specification are all prior art known to those skilled in the art.

[0026] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A method for dual-mode fault diagnosis of a tunnel electro-mechanical device, characterized in that, Includes the following steps: (a) The real-time acquired tunnel motor equipment operation data is preprocessed and feature extracted to analyze its short-term dynamics. Time-domain / instantaneous correlation features are extracted to generate real-time fault feature vectors corresponding to the instantaneous state of the equipment. Frequency domain transformation and sliding window method are applied to the operation data to extract historical sequence features and generate trend fault feature vectors. The operation data includes vibration signals, temperature signals and current signals. (b) Input the real-time fault vector into the first diagnostic model, obtain the corresponding fault judgment through threshold comparison, and generate the first fault diagnosis result; (c) Input the trend fault feature vector into the second diagnostic model, identify its hidden feature faults based on the feature mapping relationship trained in history, and generate the second fault diagnosis result; (d) Based on the first fault diagnosis result and the second fault diagnosis result, a comprehensive judgment is made according to the preset fusion rules to obtain the corresponding fault type and generate a comprehensive fault diagnosis report; The fusion rules include: if the first fault diagnosis result indicates an emergency fault, then the result is directly adopted as the comprehensive judgment result; otherwise, the first fault diagnosis result and the second fault diagnosis result are weighted and fused to generate the comprehensive fault diagnosis report. (e) Based on the generated comprehensive fault diagnosis report, trigger the corresponding alarm and match the fault knowledge base to output the corresponding maintenance suggestions.

2. The method according to claim 1, wherein, The preprocessing and feature extraction in step (a) include: The vibration signal is filtered and denoised, and the temperature and current signals are standardized to eliminate the dimensional differences of environmental noise. The peak factor Cf and waveform factor Wf of the vibration signal, the real-time effective value Ir and fluctuation energy ΔIr of the current signal, and the instantaneous rate of change ΔTr of the temperature signal are extracted and combined to form a short-term dynamic feature set Sf. The short-term dynamic feature set Sf is then integrated in sequence to generate a real-time fault feature vector Vr. The vibration signal is frequency-domain transformed to extract the low-frequency energy ratio Ef, the temperature signal is analyzed by sliding window to extract the mean Tavg and the baseline offset ΔTbase, and the harmonic distortion rate mean THDavg is extracted from the current signal. These are combined to form a long-term trend feature set Lf. The long-term trend feature set Lf is then integrated in sequence to generate a trend fault feature vector Vt.

3. The method of claim 1, wherein the method further comprises: In step (b), the rule-based first diagnostic model specifically performs the following determination: Extract the effective value of current Ir from the real-time fault feature vector Vt, and compare it with the safe current threshold dynamically configured according to the equipment type and operating conditions. If the current exceeds the limit for a first preset time, it is determined to be an overload fault. The vibration peak factor Cf in the real-time fault feature vector Vt is extracted and compared with a preset dynamic threshold. If it continues to exceed the limit for a second preset duration, it is determined to be a mechanical loosening fault. The temperature value in the real-time fault feature vector Vt is extracted and compared with the preset safe temperature range. If the temperature exceeds the limit for a third preset duration, it is determined to be an overheating fault.

4. The method of claim 1, wherein the method further comprises: The second diagnostic model in step (c) is a convolutional neural network (CNN) model; The CNN model uses vibration spectrum, temperature time series and current time series extracted from equipment operation data as training samples, and uses the corresponding fault category as a label for training. The model learns the mapping relationship between the trend fault feature vector and the potential fault type to predict the potential fault type, which includes bearing wear, blade cracks and electrical faults.

5. The method of claim 1, wherein the method further comprises: In step (d), when the first fault diagnosis result does not indicate an emergency fault, the specific method of the comprehensive weighted calculation is as follows: Based on the current load change rate of the equipment, the first weight W1 of the first fault diagnosis result and the second weight W2 of the second fault diagnosis result are dynamically allocated, wherein when the load change rate is greater than a set threshold, the proportion of W1 is increased. At the same time, the corrected weight W2' is obtained by comparing the historical diagnostic accuracy of the second diagnostic model with W2. The first fault diagnosis result and the second diagnosis result are weighted and fused based on W1 and W2' to generate the comprehensive fault diagnosis result.

6. The method of claim 1, wherein, The comprehensive fault diagnosis report generated in step (e) includes the fault type, the severity level based on confidence level, and the fault evolution trend predicted based on long-term trend characteristics.

7. The method of claim 1, wherein the method further comprises: The tunnel's electromechanical equipment includes ventilation fans and lighting fixtures; The sensor deployment locations in step (a), the dynamic thresholds in step (c), and the training samples for the second diagnostic model in step (d) are all configured differently for the physical structure, operating characteristics, and typical fault modes of ventilation fans or lighting fixtures.

8. The method of claim 1, wherein the method further comprises: In step (a), vibration signals are collected by an acceleration sensor installed in the bearing of the equipment, temperature signals are collected by a temperature sensor embedded in the motor stator, and current signals are collected by a non-contact current sensor based on the Hall effect. The collected raw data is transmitted in real time to edge computing devices or cloud servers for processing via wireless communication modules or wired industrial networks.

9. The method of claim 1, wherein the method further comprises: The alarm triggering in step (e) specifically involves: If the comprehensive fault diagnosis report indicates an emergency fault, an on-site audible and visual alarm will be triggered, and a warning message will be sent to the remote monitoring center and the mobile terminals of maintenance personnel at the same time. If the indication is a potential fault, only send an early warning message to the remote monitoring center; The maintenance recommendations are generated by querying the equipment maintenance knowledge base, which stores the mapping relationship between historical fault types and verified maintenance measures.

10. A dual-mode fault diagnosis system for tunnel electro-mechanical equipment, characterized in that, The system for implementing the dual-mode fault diagnosis method for tunnel electromechanical equipment according to any one of claims 1-9 comprises: The data acquisition and transmission module is used to collect vibration, temperature and current signals through sensors deployed on the equipment and transmit the data to the computing unit in real time; The feature extraction and construction module is used to preprocess the received data and extract short-term dynamic features and long-term trend features respectively to construct real-time fault feature vectors and trend fault feature vectors. The real-time rule-based diagnosis module has a built-in first diagnosis model based on dynamic threshold rules, which is used to receive the real-time fault feature vector and output the first fault diagnosis result. The deep learning analysis module has a built-in second diagnostic model based on a convolutional neural network, which is used to receive the trend fault feature vector and output the second fault diagnosis result. The dynamic fusion decision module is used to comprehensively analyze the first and second fault diagnosis results according to preset fusion rules, and generate a comprehensive diagnosis report that includes fault type, severity and evolution trend.

Images