Fault diagnosis method of drilling machine variable frequency driving motor based on IMSOA-MCNN-BIGRU

Abstract

The invention relates to the technical field of motor fault diagnosis, in particular to a fault diagnosis method of a drilling machine variable frequency driving motor based on IMSOA-MCNN-BIGRU, and the method comprises the steps: collecting the current data of an experiment platform motor, and adding interference to simulate the data of a real drilling machine driving motor; a population initialization strategy of an SOA algorithm is improved, opposite-reverse learning is introduced in an iteration process, and an adaptive worst resampling mechanism after stagnation monitoring is added; the size of a convolution kernel, the size of a hidden layer, an initial learning rate and L2 regularization intensity of the MCNN-BIGRU model are optimized based on IMSOA, and optimized parameters are endowed to the MCNN-BIGRU model again so as to construct an IMSOA-MCNN-BIGRU classification model; and inputting the current data into which the interference is added into the IMSOA-MCNN-BIGRU classification model to obtain a diagnosis result. According to the method, the MCNN and the BIGRU are combined and complemented, the characterization capability and the fault judgment precision of complex non-stationary signals are improved, the MCNN-BIGRU model is optimized by using the improved sea gull optimization algorithm, and the performance of the model is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of motor fault diagnosis, and in particular to a fault diagnosis method for a drilling rig variable frequency drive motor based on IMSOA (Improved-Seagull Optimization Algorithm)-MCNN (Multi-column Convolutional Neural Network)-BIGRU (Bidirectional Gated Recurrent Unit). Background Art

[0002] In geological exploration and coal mining operations, drilling rigs serve as core equipment, and their operating efficiency and stability directly impact the safety and economic viability of the entire production system. The drive motor is the drill rig's key power source, and its operating status plays a decisive role in its ability to operate continuously during the drilling process. Hydraulic drive systems have long been widely used in drilling equipment under complex geological conditions due to their high torque output and excellent responsiveness. However, with the rapid development of motor control technology, electric motor drive has gradually become an important power option for equipment such as coal drilling rigs, demonstrating significant advantages in terms of green energy conservation and intelligent control.

[0003] Despite this, coal drilling rigs operate in extremely harsh environments, often facing a variety of extreme conditions, including high humidity, high dust levels, drastic temperature fluctuations, continuous high-load operation, and strong shock and vibration. These conditions pose a severe challenge to the motor's operational stability. Under these conditions, the drive motor is prone to various performance degradation and failure issues, such as broken rotor bars, air gap eccentricity, inter-turn short circuits, and bearing wear. Once these failures occur, they can lead to reduced equipment efficiency and increased energy consumption at the very least, or even complete machine downtime and even safety accidents at the worst, posing a significant threat to the continuity and safety of coal mine operations.

[0004] Currently, on-site motor maintenance in coal mines still relies primarily on periodic inspections and manual judgment, resulting in low efficiency and large errors. Accurately extracting key features is particularly difficult in environments with strong interference, non-stationary conditions, and complex background signals, leading to problems such as missed detections and misjudgments. While existing support vector machine (SVM) methods can autonomously learn and perform diagnostics, SVMs rely on handcrafted features and struggle to automatically extract multi-level representations, limiting their performance on complex, nonlinear, and large-scale data. Traditional 1D-CNN (one-dimensional convolutional neural network) methods typically use a single convolution kernel size, making it difficult to simultaneously capture local, fine-grained information and long-range trends. Summary of the Invention

[0005] The technical problem to be solved by the present invention is: In order to overcome the above technical problems, the present invention provides a fault diagnosis method for a drilling rig variable frequency drive motor based on IMSOA-MCNN-BIGRU. By improving the motor fault diagnosis algorithm, the parameters of the MCNN-BIGRU model can be optimized, thereby achieving the result of improving the diagnostic accuracy.

[0006] Multi-scale convolutional neural networks (MCNNs) stem from the need to extract multi-scale time-frequency features from one-dimensional signals. MCNNs deploy multiple sets of convolution kernels of different lengths in parallel, or stack them within the same layer, to obtain feature representations with different receptive fields in a single forward pass. The Bidirectional Gated Recurrent Unit (BiGRU) provides a solution to the problem of insufficient temporal correlation. The multi-scale channel features extracted by the MCNN are fed into the BiGRU, which further filters discriminative features within the global context of the time domain. The performance of the MCNN-BIGRU model is influenced by several key parameters, such as kernel size, hidden layer size, initial learning rate, and L2 regularization strength. Therefore, to improve the performance of the MCNN-BiGRU model, its parameters must be optimized; however, conventional optimization methods often tend to fall into local optima and have limited convergence accuracy.

[0007] The technical solution adopted by the present invention to solve its technical problems is: a fault diagnosis method for a drilling rig variable frequency drive motor based on IMSOA-MCNN-BIGRU, which uses a multi-scale convolutional network to extract fine-grained local features and long time window features of the signal, then uses BIGRU to simultaneously model the temporal dependencies of the sequence features output by the MCNN in a forward and backward manner, integrates context information, highlights the long-range dynamic patterns related to the fault, and constructs an MCNN-BIGRU model; uses an improved seagull optimization algorithm to optimize the convolution kernel size, hidden layer size, initial learning rate, and L2 regularization strength of the MCNN-BIGRU model to obtain an IMSOA-MCNN-BIGRU classification model, which specifically includes the following steps: Step S1: Building a faulty motor experimental platform and collecting current data of the motor under various states from the faulty motor experimental platform, and adding interference to make it closer to the drilling rig data in a real environment; Step S2: Improve the population initialization method of the SOA algorithm, add a global search strategy and a population restart strategy, introduce adversarial-reverse learning in the iterative process, and add an adaptive worst-case resampling mechanism after stagnation monitoring; Step S3: First, a multi-scale convolutional neural network is used to extract local time-frequency features under different receptive fields to enhance the characterization capability of various fault modes. Then, a bidirectional gated recurrent unit (BiGRU) is used to model temporal dependencies to capture the contextual information of the signal in the temporal dimension, thereby achieving more accurate fault identification. An MCNN-BIGRU model is constructed. Based on the improved SOA algorithm (IMSOA) in step 2, the convolution kernel size, hidden layer size, initial learning rate, and L2 regularization strength of the MCNN-BIGRU model are optimized. The optimized parameters are then re-assigned to the MCNN-BIGRU model to construct the IMSOA-MCNN-BIGRU classification model. Step S4: input the current data after adding interference in step S1 into the IMSOA-MCNN-BIGRU classification model in step S3 to obtain the motor fault diagnosis result corresponding to the motor current data.

[0008] The step S2 comprises the following steps: S2-1, use LHS-Tent mixed sampling to initialize the population, and perform an offline adversarial-reverse learning to select the POP individual with the best fitness from the 2*POP individuals as the initial population; S2-2, calculate the fitness of the individual seagull and record the current global optimal position; S2-3, update the shrinkage factor according to the linear decay formula as the number of iterations increases; S2-4, generate a new position vector for each seagull using the circling-swooping spiral model; S2-5, perform online adversarial-reverse learning on the new population and do boundary checking; S2-6, re-evaluate fitness and refresh the global optimal position; S2-7. If there is no improvement in the global optimal solution for N consecutive generations, replace the worst solution in the population with the LHS-Tent random solution. P% Individuals to achieve population restart; S2-8. If the number of iterations reaches the upper limit, the optimal value and the convergence curve are output; otherwise, the process returns to step S2-3 to continue the iteration.

[0009] Therefore, the LHS-Tent hybrid initialization is introduced in Seagull to provide a uniform and diverse first-generation population, and the iterative level adversarial-reverse learning is used to continuously expand the search space, and the worst after stagnation detection is P% Resampling injects new vitality when convergence stagnates, and the three work together to significantly improve the global search capability and final convergence accuracy.

[0010] In step S2-1, the mapping formula of Tent is:

[0011] The formula for LHS-Tent to generate population is:

[0012] in, x k is the chaotic variable of the current iteration, x n+1 is the output value at the next moment, P ij Indicates the j Individuals in i Random permutation values ​​obtained by Latin hypercube sampling on the dimension; N represents the population size; a Indicates the proportion of individuals initialized by LHS in the total population; L i , U i Respectively represent i The lower and upper bounds of the dimensional variable; x n Represents the chaotic map n The sequence value generated by the iteration; r Represents the control parameters of the chaotic map.

[0013] Before the fitness evaluation in steps S2-1 and S2-5, generate the opposite solution for each individual in the current population. , and its generating formula is

[0014] in, X i,j Indicates the current population i Individuals in j The value of the dimension, and are the lower and upper bound vectors of the search space respectively; and the original individual and the opposing solution are merged to participate in the fitness ranking.

[0015] In step S2-3, the shrinkage factor With the number of iterations Linearly decreasing, the calculation formula is

[0016] in is the current iteration number, is the maximum number of iterations, are the initial value and final value of the shrinkage factor, respectively. is the contraction factor of this generation; And based on the shrinkage factor, generate coefficient vectors A and C: ,

[0017] in, is a uniform random vector of the same dimension as the decision variable; Generate the spiral radius vector B:

[0018] Where ⊙ is element-wise multiplication, is a scalar random factor.

[0019] In step S2-4, the individual seagull position is updated using a circling-swooping spiral model, with the following formula:

[0020] The symbols Indicates multiplication by dimension, is the current global optimal position, Indicates the The current position vector of each individual seagull, is the updated position vector, A and C are coefficient vectors generated based on the shrinkage factor.

[0021] In step S2-7, if the global optimal fitness has not improved in N=5 consecutive iterations, p=30% individuals with the worst fitness in the current population are selected, and new individuals are regenerated by mixed sampling using the LHS-Tent population generation formula for replacement, and the stagnation counter is cleared.

[0022] In step S1, current data of the motor under five states are collected from the faulty motor experimental platform.

[0023] The collected current data of the motor under five states include: normal current data, current data of broken rotor bars, current data of static eccentricity of 0.139 mm, current data of static eccentricity of 0.274 mm, and current data of static eccentricity of 0.4 mm.

[0024] Four disturbance enhancement strategies were applied to the motor current data under five conditions: Gaussian white noise, low-frequency harmonic perturbations, linear drift trend, and pulse perturbations. These perturbations correspond to simulated electromagnetic noise, structural vibration, sensor stability, and transient impact events, respectively. The perturbation intensities were set as follows: white noise standard deviation of 0.02, low-frequency perturbation amplitude of 0.1, maximum drift amplitude of 0.05, pulse impact intensity of 1, and frequency of 10.

[0025] Compared with the prior art, the present invention has the following beneficial effects: 1. When initializing the population of the Seagull optimization algorithm, the present invention ensures that each dimension is uniformly sampled once and only once through Latin hypercube stratification, making the first-generation population more evenly distributed and more diverse in high-dimensional space, reducing the probability of early convergence to an inferior solution; at the same time, the Tent mapping chaotic perturbation is used to introduce high-frequency perturbations while maintaining LHS uniformity, making the population more dispersed and the differences between individuals greater, thereby increasing the probability of escaping local extreme values.

[0026] 2. The present invention inserts iterative adversarial-reverse learning between the new position generation and fitness evaluation of the Seagull optimization algorithm. It dynamically expands the search area for each generation, which is equivalent to synchronous detection on the opposite side of the optimal direction. This not only improves the global exploration diversity, but also provides the population with the opportunity to quickly jump out of the local extreme value, thereby accelerating convergence and reducing the risk of falling into the local optimum early.

[0027] 3. After each generation of the Seagull Optimization Algorithm completes fitness evaluation and updates the global optimal value, the present invention inserts a stagnation check and worst-case p% resampling. If the optimal value of each new generation for N consecutive generations is not better than the historical optimal value, the search is determined to have stalled. At this time, the worst-case p% individuals are extracted from the current population by fitness sorting, and these are replaced by new random solutions generated using LHS-Tent hybrid sampling, and the stagnation counter is reset. This method "kicks out" low-quality individuals when convergence stalls and injects a new population with high diversity. This not only breaks the constraints of local extreme values, but also maintains population activity, significantly improving the global search capability and final convergence accuracy of the improved SOA.

[0028] 4. The present invention uses the improved Seagull optimization algorithm to optimize the parameters of the MCNN-BIGRU model, such as the convolution kernel size, hidden layer size, initial learning rate and L2 regularization strength, so that the fault diagnosis method shows stronger accuracy and robustness.

[0029] 5. The present invention combines MCNN with BIGRU, so that the model can not only synchronously capture the fine-grained high-frequency details and long-window low-frequency trends of the signal in the convolution stage, but also vertically integrate the contextual dependencies of the previous and next moments in the loop stage, thereby improving the characterization ability of complex non-stationary signals and the accuracy of fault identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] The present invention will be further described below with reference to the accompanying drawings and examples.

[0031] Figure 1 Schematic diagram of the structure of the fault diagnosis method of the drilling rig variable frequency drive motor based on IMSOA-MCNN-BIGRU of the present invention; Figure 2 A comparison chart of the test results of the IMSOA algorithm of the present invention and the existing algorithm; Figure 3This is the fault result classification diagram of MCNN-BIGRU; Figure 4 This is the fault result classification diagram of SOA-MCNN-BIGRU; Figure 5 This is the fault result classification diagram of IMSOA-MCNN; Figure 6 This is the fault result classification diagram of the IMSOA-MCNN-BIGRU of the present invention. DETAILED DESCRIPTION

[0032] The present invention will now be described in further detail with reference to the accompanying drawings. It should be emphasized that the following description is merely exemplary and is not intended to limit the scope of the present invention and its application.

[0033] like Figure 1 FIG. 1 is a preferred embodiment of the present invention. The fault diagnosis method of the drilling rig variable frequency drive motor based on IMSOA-MCNN-BIGRU in this embodiment includes the following steps: S1. Build a faulty motor experimental platform and collect current data of the motor under five conditions from the faulty motor experimental platform, and add interference to make it closer to the drilling rig data in a real environment; The experimental platform primarily consists of a variable frequency drive unit, a three-phase variable frequency asynchronous motor, a load system, and a data acquisition module. It can simulate the motor's operating characteristics and fault conditions under different operating conditions. The motor used in this experimental platform is the YXVF80M1-2, a three-phase variable frequency asynchronous motor with a rated speed of 2855 rpm, a rated power of 0.75 kW, and a pole pair number of p = 2. Broken rotor bars are calibrated by drilling holes in the internal rotor copper bars. To adjust the static eccentricity of the motor, dials are added to the front and rear of the experimental motor. The experimental platform diagnoses five types of motor faults: normal motor, broken rotor bar motor, motor with a static eccentricity of 0.139 mm, motor with a static eccentricity of 0.274 mm, and motor with a static eccentricity of 0.4 mm.

[0034] S2. Improve the population initialization strategy of the SOA algorithm, introduce adversarial-reverse learning in the iteration process, and add an adaptive worst p% resampling mechanism after stagnation monitoring; S3. First, a multi-scale convolutional neural network is used to extract local time-frequency features under different receptive fields to enhance the characterization of various fault modes. Then, a bidirectional gated recurrent unit (BiGRU) is used to model temporal dependencies to capture the contextual information of the signal in the temporal dimension, thereby achieving more accurate fault identification. An MCNN-BIGRU model is constructed. Based on the IMSOA in step 2, the convolution kernel size, hidden layer size, initial learning rate, and L2 regularization strength of the MCNN-BIGRU model are optimized. The optimized parameters are then re-assigned to the MCNN-BIGRU model to construct the IMSOA-MCNN-BIGRU classification model. S4. Input the current data after adding interference in step S1 into the IMSOA-MCNN-BIGRU classification model in step S3 to obtain a motor fault diagnosis result corresponding to the motor current data.

[0035] Therefore, the present invention uses a multi-scale convolutional network to extract fine-grained local features and longer time window features of the signal, and then uses BIGRU to simultaneously model the temporal dependencies of the sequence features output by MCNN from forward and backward, integrate contextual information, and highlight the long-range dynamic patterns related to faults; the improved Seagull optimization algorithm is used to optimize the parameters of the MCNN-BIGRU model, such as the convolution kernel size, hidden layer size, initial learning rate and L2 regularization strength, so that the fault diagnosis method can show stronger accuracy and robustness.

[0036] Specifically, in step S1, the motor used in the experimental platform is model YXVF80M1-2, a variable-frequency, speed-adjustable, three-phase asynchronous motor with a rated power of 0.75 kW. The speed is set at 2500 rpm, and current data of the drive motor is collected under five different states.

[0037] In this embodiment, step S2 includes the following steps: S2-1, use LHS-Tent mixed sampling to initialize the population, and perform an offline adversarial-reverse learning to select the POP individual with the best fitness from the 2*POP individuals as the initial population; S2-2, calculate the fitness of the individual seagull and record the current global optimal position; S2-3, update the shrinkage factor according to the linear decay formula as the number of iterations increases; S2-4, generate a new position vector for each seagull using the circling-swooping spiral model; S2-5, perform online adversarial-reverse learning on the new population and do boundary checking; S2-6, re-evaluate fitness and refresh the global optimal position; S2-7. If there is no improvement in the global optimal solution for N consecutive generations, replace the worst solution in the population with the LHS-Tent random solution. P% Individuals to achieve population restart; S2-8. If the number of iterations reaches the upper limit, the optimal value and the convergence curve are output; otherwise, the process returns to step S2-3 to continue the iteration.

[0038] In step S2-1, the mapping formula of Tent is:

[0039] The formula for LHS-Tent to generate population is:

[0040] in, x k is the chaotic variable of the current iteration, x n+1 is the output value at the next moment, Indicates the Individuals in Random permutation values ​​obtained by Latin hypercube sampling on the dimension; represents the population size; Indicates the proportion of individuals initialized by LHS in the total population; , Respectively represent The lower and upper bounds of the dimensional variable; Represents the chaotic map The sequence value generated by the iteration; Represents the control parameters of the chaotic map.

[0041] Before the fitness evaluation in steps S2-1 and S2-5, generate the opposite solution for each individual in the current population. , and its generating formula is

[0042] in, X i,j Indicates the current population i Individuals in j The value of the dimension, and are the lower and upper bound vectors of the search space respectively; and the original individual and the opposing solution are merged to participate in the fitness ranking.

[0043] The shrinkage factor in step S2-3 With the number of iterations Linearly decreasing, the calculation formula is

[0044] in is the current iteration number, is the maximum number of iterations, are the initial value and final value of the shrinkage factor, respectively. is the shrinkage factor of this generation.

[0045] And based on the shrinkage factor, generate coefficient vectors A and C: ,

[0046] in, is a uniform random vector of the same dimension as the decision variable.

[0047] Generate the spiral radius vector B:

[0048] Where ⊙ is element-wise multiplication, is a scalar random factor.

[0049] Furthermore, in step S2-4, the individual seagull position is updated using a circling-swooping spiral model, and the formula is as follows:

[0050] The symbols Indicates multiplication by dimension, is the current global optimal position, Indicates the The current position vector of each individual seagull, is the updated position vector, and A and C are the coefficient vectors generated according to claim 5.

[0051] In step S2-7, if the global optimal fitness has not improved in N=5 consecutive iterations, p=30% individuals with the worst fitness in the current population are selected, and new individuals are regenerated by mixed sampling using the LHS-Tent population generation formula to replace them, and the stagnation counter is cleared.

[0052] Therefore, LHS-Tent hybrid initialization is introduced into the Seagull optimization algorithm to provide a uniform and diverse first-generation population, iterative adversarial-inverse learning is used to continuously expand the search space, and the worst p% resampling after stagnation detection injects new vitality when convergence stagnates. The three together significantly improve the global search power and final convergence accuracy.

[0053] Specifically, the parameter settings of the improved SOA algorithm are shown in Table 1.

[0054] Table 1: Parameters of the improved SOA algorithm

[0055] Specifically, in step S2-2, the cross entropy loss of the validation set is used as the fitness of the calculation algorithm, and the calculation formula is:

[0056] in, Represents the batch average cross entropy loss, that is, the fitness value; Indicates the number of samples in the validation set; Indicates the number of classification categories; Representation sample The true label on category c is in the form of one-hot. If the sample belongs to this category, ,otherwise ; Model-to-sample Belong to category The predicted probability of .

[0057] In this embodiment, in step S3, the convolution kernel size, hidden layer size, initial learning rate and L2 regularization strength of the MCNN-BIGRU model are optimized through IMSOA, the optimized parameters are re-assigned to the MCNN-BIGRU model, and the MCNN-BIGRU model with updated parameters is retrained and tested.

[0058] It should be noted that: Figure 3 As shown in the figure, to verify the effectiveness of the fault diagnosis method for the variable frequency drive motor of the drilling rig of the present invention, the designed IMSOA algorithm, SOA and SSA algorithms are optimized for the same test function, with the race size set to 20 and the number of iterations to 200. In order to see the performance comparison more intuitively, the fitness value is log10 transformed, and it can be seen that the optimization and convergence capabilities of the IMSOA algorithm are better than those of SOA and SSA. The motor used in the experimental platform is the YXVF80M1-2, a variable-frequency, speed-regulating, three-phase asynchronous motor with a rated power of 0.75 kW. The speed was set at 2500 rpm, and current data from the drive motor was collected under five different conditions.

[0059] The labels for motor fault types are shown in Table 2; Table 2: Classification of motor fault types

[0060] like Figures 3 to 6As shown in the figure, the classification results of the test set using the models MCNN-BIGRU, SOA-MCNN-BIGRU, IMSOA-MCNN, and IMSOA-MCNN-BIGRU are respectively. The classification accuracy of the method of the present invention, i.e., IMSOA-MCNN-BIGRU, is 99.79%, while the classification results of MCNN-BIGRU, SOA-MCNN-BIGRU, and IMSOA-MCNN are 96.35%, 97.21%, and 91.20%, respectively. Therefore, the model of the present invention has better superiority.

[0061] In summary, this invention combines the powerful advantages of the Improved Seagull Optimization Algorithm (IMSOA) with the multi-scale convolutional neural network (MCNN) and bidirectional GRU (BiGRU) for motor fault diagnosis. IMSOA optimizes the hyperparameters of the MCNN-BiGRU model by introducing techniques such as LHS-Tent hybrid initialization, iterative level adversarial-backward learning (OBL), and stagnation detection resampling, thereby improving the model's global search capability and accuracy in high-dimensional, complex data. MCNN extracts multi-scale features of the input signal, while BiGRU effectively captures long-term and short-term dependencies in time series through its bidirectional learning mechanism. Combined with IMSOA for hyperparameter optimization, the model demonstrates superior fault diagnosis capabilities and faster convergence than traditional optimization algorithms, significantly improving fault classification accuracy.

[0062] With the above-described preferred embodiments of the present invention as a guide, and with reference to the above description, relevant personnel are fully capable of making various changes and modifications without departing from the technical scope of this invention. The technical scope of this invention is not limited to the contents of the specification and must be determined according to the scope of the claims.

Claims

1. A fault diagnosis method for a drilling rig variable frequency drive motor based on IMSOA-MCNN-BIGRU, characterized in that: A multi-scale convolutional network is used to extract fine-grained local features and longer time window features of the signal. A Big Integral Regression Unit (BIGRU) is then used to simultaneously model the temporal dependencies of the sequence features output by the MCNN in both the forward and backward directions. This integrates contextual information, highlights long-range dynamic patterns related to faults, and constructs an MCNN-BIGRU model. The improved Seagull optimization algorithm is used to optimize the convolution kernel size, hidden layer size, initial learning rate, and L2 regularization strength of the MCNN-BIGRU model to obtain the IMSOA-MCNN-BIGRU classification model. This model specifically includes the following steps: Step S1: Building a faulty motor experimental platform and collecting current data of the motor under various states from the faulty motor experimental platform, and adding interference to make it closer to the drilling rig data in a real environment; Step S2: Improve the population initialization method of the SOA algorithm, add a global search strategy and a population restart strategy, introduce adversarial-reverse learning in the iterative process, and add an adaptive worst-case resampling mechanism after stagnation monitoring; Step S3: First, a multi-scale convolutional neural network is used to extract local time-frequency features under different receptive fields. Then, the temporal dependency is modeled through a bidirectional gated recurrent unit to capture the contextual information of the signal in the time dimension. The MCNN-BIGRU model is constructed. Based on the improved SOA algorithm in step S2, the convolution kernel size, hidden layer size, initial learning rate, and L2 regularization strength of the MCNN-BIGRU model are optimized. The optimized parameters are re-assigned to the MCNN-BIGRU model to construct the IMSOA-MCNN-BIGRU classification model. Step S4: input the current data after adding interference in step S1 into the IMSOA-MCNN-BIGRU classification model in step S3 to obtain the motor fault diagnosis result corresponding to the motor current data.

2. The fault diagnosis method for a drilling rig variable frequency drive motor based on IMSOA-MCNN-BIGRU according to claim 1, characterized in that: The step S2 comprises the following steps: S2-1, use LHS-Tent mixed sampling to initialize the population, and perform an offline adversarial-reverse learning to select the POP individual with the best fitness from the 2*POP individuals as the initial population; S2-2, calculate the fitness of the individual seagull and record the current global optimal position; S2-3, update the shrinkage factor according to the linear decay formula as the number of iterations increases; S2-4, generate a new position vector for each seagull using the circling-swooping spiral model; S2-5, perform online adversarial-reverse learning on the new population and do boundary checking; S2-6, re-evaluate fitness and refresh the global optimal position; S2-7. If there is no improvement in the global optimal solution for N consecutive generations, replace the worst solution in the population with the LHS-Tent random solution. P% Individuals to achieve population restart; S2-8. If the number of iterations reaches the upper limit, the optimal value and the convergence curve are output; otherwise, the process returns to step S2-3 to continue the iteration.

3. The fault diagnosis method for a drilling rig variable frequency drive motor based on IMSOA-MCNN-BIGRU according to claim 2, characterized in that: In step S2-1, the mapping formula of Tent is: The formula for LHS-Tent to generate population is: in, x k is the chaotic variable of the current iteration, x n+1 is the output value at the next moment, P ij Indicates the j Individuals in i Random permutation values ​​obtained by Latin hypercube sampling on the dimension; N represents the population size; a Indicates the proportion of individuals initialized by LHS in the total population; L i , U i Respectively represent i The lower and upper bounds of the dimensional variable; x n Represents the chaotic map n The sequence value generated by the iteration; r Represents the control parameters of the chaotic map.

4. The fault diagnosis method for a drilling rig variable frequency drive motor based on IMSOA-MCNN-BIGRU according to claim 1 or 3, characterized in that: Before the fitness evaluation in steps S2-1 and S2-5, generate the opposite solution for each individual in the current population. , and its generating formula is in, X i,j Indicates the current population i Individuals in j The value of the dimension, and are the lower and upper bound vectors of the search space respectively; and the original individual and the opposing solution are merged to participate in the fitness ranking.

5. The fault diagnosis method for a drilling rig variable frequency drive motor based on IMSOA-MCNN-BIGRU according to claim 4, characterized in that: In step S2-3, the shrinkage factor With the number of iterations Linearly decreasing, the calculation formula is in is the current iteration number, is the maximum number of iterations, are the initial value and final value of the shrinkage factor, respectively. is the contraction factor of this generation; And based on the shrinkage factor, generate coefficient vectors A and C: ， in, is a uniform random vector of the same dimension as the decision variable; Generate the spiral radius vector B: Where ⊙ is element-wise multiplication, is a scalar random factor.

6. The fault diagnosis method for a drilling rig variable frequency drive motor based on IMSOA-MCNN-BIGRU according to claim 5, characterized in that: In step S2-4, the individual seagull position is updated using a circling-swooping spiral model, with the following formula: The symbols Indicates multiplication by dimension, is the current global optimal position, Indicates the The current position vector of each individual seagull, is the updated position vector, A and C are coefficient vectors generated based on the shrinkage factor.

7. The fault diagnosis method for a drilling rig variable frequency drive motor based on IMSOA-MCNN-BIGRU according to claim 6, characterized in that: In step S2-7, if the global optimal fitness has not improved in N=5 consecutive iterations, p=30% individuals with the worst fitness in the current population are selected, and new individuals are regenerated by mixed sampling using the LHS-Tent population generation formula for replacement, and the stagnation counter is cleared.

8. The fault diagnosis method for a drilling rig variable frequency drive motor based on IMSOA-MCNN-BIGRU according to claim 1, characterized in that: In step S1, current data of the motor under five states are collected from the faulty motor experimental platform.

9. The fault diagnosis method for a drilling rig variable frequency drive motor based on IMSOA-MCNN-BIGRU according to claim 8, characterized in that: The motor current data collected in 5 states include: Normal current data, current data of broken rotor bars, current data with a static eccentricity of 0.139 mm, current data with a static eccentricity of 0.274 mm, and current data with a static eccentricity of 0.4 mm.

10. The fault diagnosis method for a drilling rig variable frequency drive motor based on IMSOA-MCNN-BIGRU according to claim 9, characterized in that: Four disturbance enhancement strategies are added based on the current data of the motor under five states, including Gaussian white noise, low-frequency harmonic disturbance, linear drift trend and pulse disturbance.

Images