Coupler intelligent diagnosis method and system

Abstract

The invention discloses a coupling intelligent diagnosis method and system, and the method comprises the steps: constructing a decision tree-lightweight Transform model in an equipment operation state initial judgment module, combining real-time collected data and historical fault features, outputting a health level, and carrying out the pre-judgment of the collected data; when the health level is lower than a preset threshold value, the deep learning intelligent diagnosis module performs intelligent diagnosis on the preprocessed data, and the deep learning intelligent diagnosis module optimizes a Q-Former model hyper-parameter through a starfish optimization algorithm with a self-adaptive search step length to realize diagnosis. The intelligent diagnosis result output module outputs a chart form for four-level early warning; and a good communication mode is preferentially selected through the communication module based on the current network condition, and result information is transmitted to the user side. Compared with the prior art, intelligent diagnosis and early warning of the coupler can be achieved, and the stability and accuracy of vehicle operation are effectively improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent diagnosis of couplings, and in particular relates to an intelligent diagnosis method and system for couplings. Background Art

[0002] As a core component of industrial machinery transmission systems, couplings are widely used in equipment such as fans, pumps, and generators, fulfilling a crucial role in transmitting power, compensating for shaft misalignment, and buffering impact loads. With the rapid development of the automotive industry toward electrification and intelligentization in recent years, the application of couplings in new energy vehicles has significantly expanded, becoming a key connection between the drive motor and the transmission system. The increasing popularity of electric vehicles has driven the development of high-speed, high-precision vehicles. The operational stability and reliability of couplings directly determine the overall performance and lifespan of the vehicle. In electric vehicles, couplings must withstand the high frequency of starts and stops, the instantaneous high torque output, and the reverse load shocks during energy recovery. Furthermore, hybrid vehicles require seamless power transfer between the internal combustion engine and the electric motor, placing even higher demands on the coupling's dynamic response accuracy and fatigue life. Furthermore, all vehicles are prone to heavy load shock, high temperatures, and frequent starts and stops during operation. These conditions can easily lead to misalignment, wear, and fatigue cracks, which can cause equipment downtime or even accidents.

[0003] Traditional coupling fault diagnosis methods mainly rely on manual experience and single-point threshold detection, and make preliminary judgments through manual inspections or threshold alarms of single parameters. There is a large load pressure for large-scale data transmission, and the false alarm rate is high. In addition, couplings of various brands have structural differences, resulting in fragmented diagnostic rules and extremely high reliance on experienced engineers. They are lagging, one-sided, inefficient and difficult to promote on a large scale. Traditional diagnostic systems mostly use a single communication protocol or simple binary switching based on signal strength and LoRa protocol, lack dynamic optimization capabilities, resulting in high resource usage, and protocol switching relies on threshold judgment. There is a risk of transmission interruption when switching in weak scenarios, there is no priority guarantee, and emergency faults and regular data are mixed in transmission, which delays maintenance opportunities. Traditional coupling fault diagnosis mostly uses fixed-parameter machine learning algorithms, relies on manual feature engineering, and model updates require retraining. The calculation data is large, which increases the system load and cannot meet real-time requirements.

[0004] Therefore, a smart coupling diagnosis method is needed that can handle complex operating conditions and improve diagnostic accuracy. This method, combined with a module for initially determining the equipment's operating status and a preferred communication method, can improve the efficiency and stability of data transmission. This method should comprehensively consider the entire diagnostic system process, enabling rapid and accurate diagnosis of coupling faults and providing strong support for the safe and stable operation of the vehicle. Summary of the Invention

[0005] Purpose of the invention: The present invention provides an intelligent diagnosis method and system for a coupling to realize intelligent diagnosis of the coupling.

[0006] Technical solution: The present invention discloses an intelligent diagnosis method for a coupling, comprising the following steps:

[0007] Step 1: real-time collection of coupling operation data, including but not limited to speed, temperature, voltage, current and rotational position, and at the same time, collection of historical coupling fault information to build a historical fault database and statistically analyze historical fault characteristics;

[0008] Step 2: Preprocess the real-time collected coupling operation data, including interpolation, outlier removal, and normalization;

[0009] Step 3: Combining the historical fault database and real-time coupling operation data, a decision tree-lightweight Transformer model is constructed. A scaled dot-product attention mechanism is used to dynamically weight the model to generate a comprehensive health level. The health level is then compared with a preset threshold. If the health level falls below the threshold, the preprocessed coupling operation data is passed to step 4 for further diagnosis.

[0010] Step 4: Build a Q-Former intelligent diagnosis model. Use the preprocessed operating data as the input of the Q-Former intelligent diagnosis model. Design an objective function that considers both diagnostic accuracy and model robustness. Minimize the objective function value as the model parameter optimization direction. Use the Starfish optimization algorithm to optimize the Q-Former intelligent diagnosis model parameters to achieve intelligent diagnosis of the coupling working status and identify faults.

[0011] Furthermore, the specific steps of step 3 are as follows:

[0012] (1) The feature extraction module performs dual feature extraction, extracts trend features from real-time coupling operation data, and calculates the probability of failure modes from the confusion matrix of the historical fault library;

[0013] (2) Use the attention mechanism to perform linear transformation on the preprocessed coupling operation data and historical fault features:

[0014] Q=G t W Q ,K=MW K ,V=MW V

[0015] Where W Q ,W K ,W V is a learnable parameter matrix that maps the original features to the query Q, key K, value V space;

[0016] (3) Calculate the correlation between real-time data and historical failure modes by scaling dot product attention:

[0017]

[0018] Where C is the attention weight, d K is the dimension of the key vector, used to prevent gradient explosion;

[0019] (4) Apply the attention weight C to the historical fault feature value V to generate a weighted vector, and finally output O t The dynamic correlation characteristics between current real-time data and historical failure modes are expressed as follows:

[0020] O t =C·V

[0021] (5) The real-time sequence feature G t With attention output O t Splicing, get the fusion feature F = [G t ;O t ];

[0022] (6) The health level HL is obtained through the full connection layer and nonlinear activation function mapping:

[0023] HL=σ(W h F+b h )

[0024] Where W h ,b h is the fully connected layer parameter, σ is the Sigmoid function, and the output range is [0,1].

[0025] Furthermore, the starfish optimization algorithm is used to optimize the parameters of the Q-Former intelligent diagnosis model. Specifically, the starfish optimization algorithm SFOA is used to optimize the hyperparameters of the Q-Former intelligent diagnosis model, where the hyperparameters of the Q-Former intelligent diagnosis model include but are not limited to learning rate, batch size, query length and hidden layer size.

[0026] Furthermore, the objective function calculation formula is:

[0027] Obj=ω1FAR+ω2MAR+ω3max(MAR Train ,MAR Test ,MAR V )

[0028] Where, FAR is the false alarm rate of the Q-Former intelligent diagnosis model, MAR is the average missed alarm rate of the Q-Former intelligent diagnosis model, and MAR Train 、MAR Testand MAR V is the training set missed rate, test set missed rate and inspection set missed rate of the Q-Former intelligent diagnosis model, ω1+ω2+ω3=1, ω b is the weight, b=1,2,3.

[0029] Furthermore, the Starfish optimization algorithm is used to optimize the parameters of the Q-Former intelligent diagnosis model. The specific steps are as follows:

[0030] (1) Random initialization, the formula is as follows:

[0031] X ij =l j +r(u j -l j )i=1,2,...,N j=1,2,...,D

[0032] Where, X ij is the j-th dimension update value of the i-th group of Q-Former intelligent diagnosis model hyperparameters, r is a random number between (0,1), u j and l j Update the upper and lower bounds of the value for the jth dimension;

[0033] (2) First determine the dimension of the hyperparameter. When the hyperparameter that needs to be optimized for the Q-Former intelligent diagnosis model is greater than 5, a five-dimensional search mode is performed:

[0034]

[0035] Where, and is the updated value and current value of the i-th hyperparameter group, is the current optimal value of the p-th dimension, p is a randomly selected dimension in D, T is the current number of iterations, T max is the maximum number of iterations, λ is the adaptive adjustment coefficient of the search step size;

[0036] When the hyperparameter dimension that needs to be optimized for the Q-Former intelligent diagnosis model is no more than 5, a one-dimensional search mode is used to update the hyperparameter group. The information of other hyperparameter groups is used to search in a specified direction. The formula is as follows:

[0037]

[0038] Where, and are the values of two randomly selected hyperparameter groups in the pth dimension, A1 and A2 are random numbers between (-1, 1), E t is the information volume control parameter.

[0039] (3) When the dimensions of the hyperparameter group are determined, the development phase begins. A parallel bidirectional search strategy is adopted. The information of other hyperparameter groups and the current best hyperparameter group are used to calculate the deviation between the best hyperparameter group and other hyperparameter groups. Then, a bidirectional search strategy is used to update each hyperparameter group. The formula is as follows:

[0040]

[0041] Where, d m The distance between the 5 globally optimal hyperparameter groups and other hyperparameter groups, m p are 5 randomly selected hyperparameter groups, are the positions of 5 randomly selected hyperparameter groups, r1 and r2 are random numbers between (0,1), and d m1 and d m2 It is d m 2 randomly selected hyperparameter groups in ;

[0042] Among all hyperparameter groups, the hyperparameter group farthest from the optimal hyperparameter group may iterate very slowly during the exploration process. When this hyperparameter group iterates towards a better position, it has five new exploration directions and its update rule is modeled as follows:

[0043]

[0044] Where T is the current iteration number, T max is the maximum number of iterations, is the updated hyperparameter group position, is the hyperparameter group farthest from the optimal hyperparameter group, and N is the population size;

[0045] If the updated value of the set of hyperparameters exceeds the set range, the set of hyperparameter values will be reset. The formula is as follows:

[0046]

[0047] The present invention also discloses a coupling intelligent diagnosis system based on the above coupling intelligent diagnosis method, comprising the following modules:

[0048] Multi-dimensional data acquisition module: used to collect coupling operation data in real time, including but not limited to speed, temperature, voltage, current and rotation position, and at the same time collect historical fault information of the coupling to build a historical fault database and calculate historical fault characteristics;

[0049] Data preprocessing module: preprocesses the real-time collected coupling operation data, including interpolation, outlier removal and normalization;

[0050] Equipment operating status initial assessment module: This module combines the historical fault database and real-time coupling operating data to build a decision tree-lightweight Transformer model. This model uses a scaled dot-product attention mechanism to dynamically weight and generate a comprehensive health level. This health level is then compared with a preset threshold. If the health level falls below the threshold, the pre-processed data is passed to the deep learning intelligent diagnosis module for diagnosis.

[0051] Deep Learning Intelligent Diagnosis Module: This module builds a Q-Former intelligent diagnosis model and uses the Starfish optimization algorithm to optimize the Q-Former model parameters. The module uses preprocessed operating data as input to the Q-Former intelligent diagnosis model. An objective function is designed that considers both diagnostic accuracy and model robustness. The model parameter optimization direction is determined by minimizing the objective function value, enabling intelligent diagnosis of the coupling's operating status and fault identification.

[0052] Preferably, it also includes:

[0053] Intelligent diagnostic result output module: Provides graded warnings using charts and linear synthesis unit results, and displays diagnostic results in a variety of ways, allowing users to clearly understand the health status of the coupling and provide a basis for equipment maintenance;

[0054] Hierarchical warning module: judges the intelligent diagnosis results, divides them into four levels of alarms, and issues warnings to the user end.

[0055] Preferably, the intelligent diagnosis results display the current status and changes of the fault, and display the diagnosis results to the user in four levels of warning:

[0056] (1) Level 1 display is normal: all operating data of the coupling are within the normal range, the operation is stable, and there are no obvious abnormal signs;

[0057] (2) Level 2 warning is concern: various operating data are close to the upper or lower limit of the normal range, and there is a potential risk, but it has not yet affected the normal operation of the coupling;

[0058] (3) Level 3 warning is a warning: all operating data are out of the normal range, and the coupling has a minor fault, but it can still continue to operate for a period of time;

[0059] (4) Level 4 warning is dangerous: various operating data are seriously exceeded, the coupling failure is serious, the machine may be shut down at any time, and even a chain failure may be triggered, affecting the safe operation of the entire system.

[0060] Preferably, it also includes a communication module to collect signal strength RSSI and delay in real time, and select the optimal communication protocol from Bluetooth, LTE transmission, low-power LoRa protocol and 5G signal transmission according to the data transmission priority set by the graded warning to complete data transmission, and add a manual emergency setting device so that in case of emergency, manual intervention can be made to continuously maintain data transmission; the specific steps are as follows:

[0061] (1) Real-time acquisition of signal strength RSSI, packet loss rate, and latency, and classification of the diagnostic results to assist the decision tree model in making judgments and selections. When the diagnostic results are level 1 and level 2 warnings, the data transmission priority is determined to be low; when the diagnostic results are level 3 and level 4 warnings, the data transmission priority is determined to be high;

[0062] (2) Construct hard rules to quickly filter out infeasible communication protocols. The hard rules are as follows:

[0063] 1) When the data priority is low, RSSI>-90dBm and delay<100ms, select Bluetooth; when RSSI≤-90dBm and delay≥100ms, select LoRa protocol;

[0064] 2) When the data priority is high, RSSI>-80dBm and delay<50ms, select 5G signal transmission; when RSSI≤-80dBm and delay≥50ms, select LTE transmission;

[0065] (3) Use a lightweight Transformer model to process the RSSI and delay data collected in real time and output a priority communication protocol;

[0066] (4) Dynamically adjust the weights of the output results of the decision tree model and the lightweight Transformer model based on the success rates of the last 10 transmissions;

[0067] (5) Output weighted comprehensive results and select the optimal communication protocol;

[0068] (6) Transmit the diagnosis results and treatment suggestions to the user end.

[0069] Beneficial effects:

[0070] 1. The present invention constructs a module for preliminary judgment of equipment operation status. Through the decision tree-lightweight Transformer model, combined with real-time collected data and historical fault characteristics, the fusion output health level can realize the preliminary judgment of coupling faults, reduce data transmission pressure, and improve system operation efficiency.

[0071] The present invention constructs a starfish optimization algorithm to optimize the hyperparameters of the Q-Former model and selects the optimal hyperparameter group, which can improve the diagnostic precision and accuracy and realize intelligent diagnosis of the coupling.

[0072] The present invention constructs a communication module that, based on set hard rules, collects signal strength RSSI and delay, and can select the optimal communication solution according to the data priority set by the hierarchical warning module to achieve stable system communication. The addition of an artificial emergency device can maintain system operation and communication with the user end in an emergency. BRIEF DESCRIPTION OF THE DRAWINGS

[0073] Figure 1 Flowchart for intelligent diagnosis of coupling;

[0074] Figure 2 Flowchart for optimizing hyperparameters of the Q-Former model using the Starfish optimization algorithm. DETAILED DESCRIPTION

[0075] The present invention will be described in further detail below with reference to the accompanying drawings.

[0076] like Figure 1 As shown, the present invention provides a coupling intelligent diagnosis method and system, which realizes intelligent processing of coupling operation data and accurate identification of fault types, including the following steps:

[0077] (1) Multi-dimensional data acquisition module: used to collect the operation data of the coupling in real time, including but not limited to speed, temperature, voltage, current and rotation position, and at the same time collect the historical fault information of the coupling to build a historical fault database and count the historical fault characteristics.

[0078] (2) Data preprocessing module: preprocess the coupling operation data collected in real time, including interpolation, removal of outliers and normalization.

[0079] (3) Equipment operation status preliminary judgment module: The scaled dot product attention mechanism is used to combine the real-time collected data and historical fault characteristics, and the health level is weighted to generate the health level. The health level is compared with the preset threshold. When the health level is not higher than the preset threshold, the preprocessed data is passed to the deep learning intelligent diagnosis module for diagnosis, thereby completing the preliminary judgment of the equipment fault.

[0080] The equipment operation status initial judgment module combines the historical fault database and real-time monitoring data, and generates a comprehensive health level through dynamic weighting of the attention mechanism. If the health level does not exceed the set threshold, it is determined that the coupling is faulty and the data is sent to the deep learning intelligent diagnosis module for diagnosis. If the health level exceeds the set threshold, it is determined that the coupling is not faulty and no fault diagnosis is required. The specific steps are as follows:

[0081] (3.1) The feature extraction module performs dual feature extraction, extracting trend features from real-time data and calculating the probability of failure modes from the confusion matrix of the historical fault library;

[0082] (3.2) Use the attention mechanism to perform linear transformation on real-time sequence and historical fault features:

[0083] Q=G t W Q ,K=MW K ,V=MW V

[0084] Where W Q ,W K ,W V is a learnable parameter matrix that maps the original features to the query Q, key K, value V space.

[0085] (3.3) Calculate the correlation between real-time data and historical failure modes by scaling dot product attention:

[0086]

[0087] Where, d K is the dimension of the key vector, used to prevent gradient explosion;

[0088] (3.4) Apply the attention weight C to the historical fault feature value V to generate a weighted vector, and finally output O t The dynamic correlation characteristics between current real-time data and historical failure modes are expressed as follows:

[0089] O t =C·V

[0090] (3.5) The real-time sequence feature G t With attention output O t Splicing, get the fusion feature F = [G t ;O t ];

[0091] (3.6) The health level HL is obtained through the full connection layer and nonlinear activation function mapping:

[0092] HL=σ(W h F+b h )

[0093] Where W h ,b h is the fully connected layer parameter, σ is the Sigmoid function, and the output range is [0,1].

[0094] (4) Deep learning intelligent diagnosis module: Construct a Q-Former intelligent diagnosis model and use the starfish optimization algorithm to optimize the Q-Former model parameters; use the preprocessed operating data as the input of the Q-Former intelligent diagnosis model, design an objective function that considers both diagnostic accuracy and model robustness, and optimize the model parameters with the minimum objective function value, so as to realize intelligent diagnosis of the coupling working status and identify faults;

[0095] (4.1) The Q-Former intelligent diagnosis model in the deep learning intelligent diagnosis module is optimized using the Starfish Optimization Algorithm (SFOA) to optimize the hyperparameters of the Q-Former model. The hyperparameters of the Q-Former model include but are not limited to the learning rate, batch size, query length, and hidden layer size. The specific steps are as follows:

[0096] (4.1.1) Random initialization, the formula is as follows:

[0097] X ij =l j +r(u j -l j )i=1,2,...,N j=1,2,...,D

[0098] Where, X ij Update the j-th dimension of the i-th group of Q-Former model hyperparameters, r is a random number between (0,1), u j and l j Update the upper and lower bounds for the values of the j-th dimension.

[0099] (4.1.2) Due to technical barriers between different brands of coupling manufacturers, the hyperparameters that need to be optimized for the diagnostic models of different brands of couplings vary. This requires exploration in more dimensional hyperparameter groups, so the hyperparameter dimensions must be determined first. When the number of hyperparameters that the Q-Former intelligent diagnostic model needs to optimize is greater than 5, a five-dimensional search mode is used:

[0100]

[0101] Where, and is the updated value and current value of the i-th hyperparameter group, is the current optimal value of the p-th dimension, p is 5 randomly selected dimensions in D, T is the current number of iterations, T max is the maximum number of iterations, λ is the adaptive adjustment coefficient of the search step size;

[0102] When the hyperparameter dimension to be optimized by the Q-Former intelligent diagnosis model is no more than 5, a one-dimensional search mode is used to update the hyperparameter group. Using the information of other hyperparameter groups, a search in a specified direction is performed. The formula is as follows:

[0103]

[0104] Where, and are the values of two randomly selected hyperparameter groups in the pth dimension, A1 and A2 are random numbers between (-1, 1), p is a randomly selected dimension in D, E t is the information volume control parameter.

[0105] (4.1.3) Once the dimensions of the hyperparameter group are determined, the development phase begins. Here, a parallel bidirectional search strategy is used. Using information from other hyperparameter groups and the current optimal hyperparameter group, the deviation between the optimal hyperparameter group and the other hyperparameter groups is calculated. Then, a bidirectional search strategy is used to update each hyperparameter group. The formula is as follows:

[0106]

[0107] Where, d m The distance between the 5 globally optimal hyperparameter groups and other hyperparameter groups, m p are 5 randomly selected hyperparameter groups, are the positions of 5 randomly selected hyperparameter groups, r1 and r2 are random numbers between (0,1), and d m1 and d m2 It is d m 2 hyperparameter groups randomly selected from .

[0108] Among all hyperparameter groups, the hyperparameter group farthest from the optimal hyperparameter group may iterate very slowly during the exploration process. When this hyperparameter group iterates towards a better position, it has five new exploration directions and its update rule is modeled as follows:

[0109]

[0110] Where T is the current iteration number, T max is the maximum number of iterations, is the updated hyperparameter group position, is the hyperparameter group farthest from the optimal hyperparameter group, and N is the population size.

[0111] If the updated value of the set of hyperparameters exceeds the set range, the set of hyperparameter values is reset to avoid falling into local optimality and enhance global convergence. The formula is as follows:

[0112]

[0113] (4.2) In the deep learning intelligent diagnosis module, the objective function calculation formula is constructed as follows:

[0114] Obj=ω1FAR+ω2MAR+ω3max(MAR Train ,MAR Test ,MAR V )

[0115] Where, FAR is the false alarm rate of the Q-Former intelligent diagnosis model, MAR is the average missed alarm rate of the Q-Former intelligent diagnosis model, and MAR Train 、MAR Test and MAR V is the training set missed rate, test set missed rate and inspection set missed rate of the Q-Former intelligent diagnosis model, ω1+ω2+ω3=1, ω b is the weight, b=1,2,3

[0116] (5) Intelligent diagnosis result output module: It provides graded warnings with charts and linear synthesis unit results, and displays the diagnosis results in a diversified manner, allowing users to clearly understand the health status of the coupling and provide a basis for equipment maintenance.

[0117] (6) Hierarchical warning module: judges the intelligent diagnosis results, divides them into four levels of alarms, and issues warnings to the user end;

[0118] The hierarchical warning module displays the current status and changes of faults based on the intelligent diagnosis results, and presents the diagnosis results to users in four levels of warning:

[0119] (6.1) Level 1 display is normal: All operating data of the coupling are within the normal range, the operation is stable, and there are no obvious abnormal signs;

[0120] (6.2) Level 2 warning is "concern": various operating data are close to the upper or lower limit of the normal range, and there is a potential risk, but it has not yet affected the normal operation of the coupling;

[0121] (6.3) Level 3 warning is a warning: various operating data are out of the normal range, and the coupling has a minor fault, such as slight imbalance or misalignment, but it can still continue to operate for a period of time;

[0122] (6.4) Level 4 warning is dangerous: all operating data seriously exceed the standard, the coupling failure is serious, the machine may stop at any time, and even cause a chain failure, affecting the safe operation of the entire system.

[0123] (7) Communication module: collects signal strength RSSI and delay in real time, and selects the optimal communication protocol from Bluetooth, LTE transmission, low-power LoRa protocol and 5G signal transmission according to the data transmission priority set by the graded warning to complete data transmission, thereby improving the stability of the transmission system. In addition, a manual emergency setting device is added so that in case of emergency, manual intervention can be made to continuously maintain data transmission;

[0124] The communication module is used to transmit diagnostic results to the remote monitoring terminal in real time, enabling remote monitoring and management. It has the ability to intelligently sense the network environment. By collecting signal strength RSSI and latency in real time, it ultimately selects the optimal communication protocol among Bluetooth, LTE transmission, the low-power LoRa protocol, and 5G signal transmission to enhance signal transmission stability. The specific steps are as follows:

[0125] (7.1) Real-time acquisition of signal strength RSSI, packet loss rate, and latency; classification of diagnostic results to assist the decision tree model in making decisions and selections. When the diagnostic results are level 1 or level 2 warnings, the data transmission priority is determined to be low; when the diagnostic results are level 3 or level 4 warnings, the data transmission priority is determined to be high.

[0126] (7.2) Construct hard rules to quickly filter out infeasible communication protocols. The hard rules are as follows:

[0127] (7.2.1) When the data priority is low, RSSI>-90dBm and delay<100ms, select Bluetooth; when RSSI≤-90dBm and delay≥100ms, select LoRa protocol;

[0128] (7.2.2) When the data priority is high, RSSI>-80dBm and delay<50ms, select 5G signal transmission; when RSSI≤-80dBm and delay≥50ms, select LTE transmission;

[0129] (7.3) Use a lightweight Transformer model to process the real-time collected RSSI and delay data and output a priority communication protocol;

[0130] (7.4) Dynamically adjust the weights of the output results of the decision tree model and the lightweight Transformer model based on the success rates of the last 10 transmissions;

[0131] (7.5) Output the weighted comprehensive results and select the optimal communication protocol.

[0132] (7.6) The diagnosis results and treatment suggestions are transmitted to the user end by the communication module.

[0133] In addition, when an abnormal situation is discovered manually, the emergency button can be activated urgently. The intelligent diagnosis system will force the use of the LoRa protocol and only transmit simplified diagnosis results. The simplified diagnosis results are the abnormal operating data and the fault diagnosed by the diagnostic model.

[0134] The above embodiments are intended only to illustrate the technical concepts and features of the present invention. Their purpose is to enable those skilled in the art to understand the contents of the present invention and implement them accordingly. They are not intended to limit the scope of protection of the present invention. Any equivalent changes or modifications made in accordance with the spirit of the present invention are intended to be covered by the scope of protection of the present invention.

Claims

1. A coupling intelligent diagnosis method, characterized in that: The following steps are involved: Step 1: real-time collection of coupling operation data, including but not limited to speed, temperature, voltage, current and rotational position, and at the same time, collection of historical coupling fault information to build a historical fault database and statistically analyze historical fault characteristics; Step 2: Preprocess the real-time collected coupling operation data, including interpolation, outlier removal, and normalization; Step 3: Combining the historical fault database and real-time coupling operation data, a decision tree-lightweight Transformer model is constructed. A scaled dot-product attention mechanism is used to dynamically weight the model to generate a comprehensive health level. The health level is then compared with a preset threshold. If the health level falls below the threshold, the preprocessed coupling operation data is passed to step 4 for further diagnosis. Step 4: Build a Q-Former intelligent diagnosis model. Use the preprocessed operating data as the input of the Q-Former intelligent diagnosis model. Design an objective function that considers both diagnostic accuracy and model robustness. Minimize the objective function value as the model parameter optimization direction. Use the Starfish optimization algorithm to optimize the Q-Former intelligent diagnosis model parameters to achieve intelligent diagnosis of the coupling working status and identify faults.

2. The intelligent diagnosis method for a coupling according to claim 1, characterized in that: The specific steps of step 3 are as follows: (1) The feature extraction module performs dual feature extraction, extracts trend features from real-time coupling operation data, and calculates the probability of failure modes from the confusion matrix of the historical fault library; (2) Use the attention mechanism to perform linear transformation on the preprocessed coupling operation data and historical fault features: Q=G t W Q ,K=MW K ,V=MW V Where W Q ,W K ,W V is a learnable parameter matrix that maps the original features to the query Q, key K, value V space; (3) Calculate the correlation between real-time data and historical failure modes by scaling dot product attention: Where C is the attention weight, d K is the dimension of the key vector, used to prevent gradient explosion; (4) Apply the attention weight C to the historical fault feature value V to generate a weighted vector, and finally output O t The dynamic correlation characteristics between current real-time data and historical failure modes are expressed as follows: About t =C·V (5) The real-time sequence feature G t With attention output O t Splicing, get the fusion feature F = [G t ;O t ]; (6) The health level HL is obtained through the full connection layer and nonlinear activation function mapping: <h2 style=";text-align:left;direction:ltr">HL = σ(W<h2 style=";text-align:left;direction:ltr"> h <h2 style=";text-align:left;direction:ltr"> F+b<h2 style=";text-align:left;direction:ltr"> h <h2 style=";text-align:left;direction:ltr"> ) Where W h ,b h is the fully connected layer parameter, σ is the Sigmoid function, and the output range is [0,1].

3. The intelligent diagnosis method for a coupling according to claim 1, characterized in that: The starfish optimization algorithm is used to optimize the parameters of the Q-Former intelligent diagnosis model. Specifically, the starfish optimization algorithm SFOA is used to optimize the hyperparameters of the Q-Former intelligent diagnosis model, where the hyperparameters of the Q-Former intelligent diagnosis model include but are not limited to learning rate, batch size, query length and hidden layer size.

4. The intelligent diagnosis method for a coupling according to claim 3, characterized in that: The objective function calculation formula is: Obj=ω1FAR+ω2MAR+ω3max(MAR Train ,MAR Test ,MAR V ) Where, FAR is the false alarm rate of the Q-Former intelligent diagnosis model, MAR is the average missed alarm rate of the Q-Former intelligent diagnosis model, and MAR Train 、MAR Test and MAR V is the training set missed rate, test set missed rate and inspection set missed rate of the Q-Former intelligent diagnosis model, ω1+ω2+ω3=1, ω b is the weight, b=1,2,3.

5. The intelligent diagnosis method for a coupling according to claim 4, characterized in that: The specific steps for optimizing the parameters of the Q-Former intelligent diagnosis model using the Starfish optimization algorithm are as follows: (1) Random initialization, the formula is as follows: X ij =l j +r(u j -l j )i=1,2,...,Nj=1,2,...,D Where, X ij is the j-th dimension update value of the i-th group of Q-Former intelligent diagnosis model hyperparameters, r is a random number between (0,1), u j and l j Update the upper and lower bounds of the value for the jth dimension; (2) First determine the dimension of the hyperparameter. When the hyperparameter that needs to be optimized for the Q-Former intelligent diagnosis model is greater than 5, a five-dimensional search mode is performed: Where, and is the updated value and current value of the i-th hyperparameter group, is the current optimal value of the p-th dimension, p is a randomly selected dimension in D, T is the current number of iterations, T max is the maximum number of iterations, λ is the adaptive adjustment coefficient of the search step size; When the hyperparameter dimension that needs to be optimized for the Q-Former intelligent diagnosis model is no more than 5, a one-dimensional search mode is used to update the hyperparameter group. The information of other hyperparameter groups is used to search in a specified direction. The formula is as follows: Where, and are the values of two randomly selected hyperparameter groups in the pth dimension, A1 and A2 are random numbers between (-1, 1), E t is the information volume control parameter. (3) When the dimensions of the hyperparameter group are determined, the development phase begins. A parallel bidirectional search strategy is adopted. The information of other hyperparameter groups and the current best hyperparameter group are used to calculate the deviation between the best hyperparameter group and other hyperparameter groups. Then, a bidirectional search strategy is used to update each hyperparameter group. The formula is as follows: Where, d m The distance between the 5 globally optimal hyperparameter groups and other hyperparameter groups, m p are 5 randomly selected hyperparameter groups, are the positions of 5 randomly selected hyperparameter groups, r1 and r2 are random numbers between (0,1), and d m1 and d m2 It is d m 2 randomly selected hyperparameter groups in ; Among all hyperparameter groups, the hyperparameter group farthest from the optimal hyperparameter group may iterate very slowly during the exploration process. When this hyperparameter group iterates towards a better position, it has five new exploration directions and its update rule is modeled as follows: Where T is the current iteration number, T max is the maximum number of iterations, Y i T is the updated hyperparameter group position, is the hyperparameter group farthest from the optimal hyperparameter group, and N is the population size; If the updated value of the set of hyperparameters exceeds the set range, the set of hyperparameter values will be reset. The formula is as follows:

6. A coupling intelligent diagnosis system based on the coupling intelligent diagnosis method according to any one of claims 1 to 5, characterized in that: Includes the following modules: Multi-dimensional data acquisition module: used to collect coupling operation data in real time, including but not limited to speed, temperature, voltage, current and rotation position, and at the same time collect historical fault information of the coupling to build a historical fault database and calculate historical fault characteristics; Data preprocessing module: preprocesses the real-time collected coupling operation data, including interpolation, outlier removal and normalization; Equipment operating status initial assessment module: This module combines the historical fault database and real-time coupling operating data to build a decision tree-lightweight Transformer model. This model uses a scaled dot-product attention mechanism to dynamically weight and generate a comprehensive health level. This health level is then compared with a preset threshold. If the health level falls below the threshold, the pre-processed data is passed to the deep learning intelligent diagnosis module for diagnosis. Deep Learning Intelligent Diagnosis Module: This module builds a Q-Former intelligent diagnosis model and uses the Starfish optimization algorithm to optimize the Q-Former model parameters. The module uses preprocessed operating data as input to the Q-Former intelligent diagnosis model. An objective function is designed that considers both diagnostic accuracy and model robustness. The model parameter optimization direction is determined by minimizing the objective function value, enabling intelligent diagnosis of the coupling's operating status and fault identification.

7. The intelligent diagnosis system for coupling according to claim 6, characterized in that: Also includes: Intelligent diagnostic result output module: Provides graded warnings using charts and linear synthesis unit results, and displays diagnostic results in a variety of ways, allowing users to clearly understand the health status of the coupling and provide a basis for equipment maintenance; Hierarchical warning module: judges the intelligent diagnosis results, divides them into four levels of alarms, and issues warnings to the user end.

8. The intelligent diagnosis system for coupling according to claim 7, characterized in that: The intelligent diagnosis results show the current status and changes of the fault, and present the diagnosis results to the user in four levels of warning: (1) Level 1 display is normal: all operating data of the coupling are within the normal range, the operation is stable, and there are no obvious abnormal signs; (2) Level 2 warning is concern: various operating data are close to the upper or lower limit of the normal range, and there is a potential risk, but it has not yet affected the normal operation of the coupling; (3) Level 3 warning is a warning: all operating data are out of the normal range, and the coupling has a minor fault, but it can still continue to operate for a period of time; (4) Level 4 warning is dangerous: various operating data are seriously exceeded, the coupling failure is serious, the machine may be shut down at any time, and even a chain failure may be triggered, affecting the safe operation of the entire system.

9. The intelligent diagnosis system for coupling according to claim 6, characterized in that: It also includes a communication module that collects signal strength RSSI and latency in real time. Based on the data transmission priority set by the graded warning, it selects the optimal communication protocol among Bluetooth, LTE transmission, low-power LoRa protocol, and 5G signal transmission to complete data transmission. In addition, a manual emergency setting device is added to enable manual intervention in emergency situations to maintain data transmission. The specific steps are as follows: (1) Real-time acquisition of signal strength RSSI, packet loss rate, and latency, and classification of the diagnostic results to assist the decision tree model in making judgments and selections. When the diagnostic results are level 1 and level 2 warnings, the data transmission priority is determined to be low; when the diagnostic results are level 3 and level 4 warnings, the data transmission priority is determined to be high; (2) Construct hard rules to quickly filter out infeasible communication protocols. The hard rules are as follows: 1) When the data priority is low, RSSI>-90dBm and delay<100ms, select Bluetooth; when RSSI≤-90dBm and delay≥100ms, select LoRa protocol; 2) When the data priority is high, RSSI>-80dBm and delay<50ms, select 5G signal transmission; when RSSI≤-80dBm and delay≥50ms, select LTE transmission; (3) Use a lightweight Transformer model to process the RSSI and delay data collected in real time and output a priority communication protocol; (4) Dynamically adjust the weights of the output results of the decision tree model and the lightweight Transformer model based on the success rates of the last 10 transmissions; (5) Output weighted comprehensive results and select the optimal communication protocol; (6) Transmit the diagnosis results and treatment suggestions to the user end.

Images