Spinning machine fault detection method and system based on machine learning

Abstract

The invention provides a spinning machine fault detection method and system based on machine learning, and relates to the technical field of fault detection.The method comprises the steps that in the running process of a spinning machine, process parameter data and result parameter data are collected and preprocessed to obtain comprehensive processing data, and fault sensitive features are extracted from the comprehensive processing data; and constructing a support vector machine model, and optimizing hyper-parameters of the support vector machine model according to a Genethink shark optimization algorithm to obtain an enhanced vector machine model. And inputting the fault sensitive features into the reinforcement vector machine model, outputting a fault prediction probability, and judging whether a preset fault judgment threshold value is reached or not, and if so, positioning a fault source according to the current fault sensitive features. And formulating a maintenance scheme according to the fault source, and regularly maintaining the spinning machine according to the fault detection interval. According to the method, the fault prediction precision can be improved, the error and omission ratio is reduced, the fault detection timeliness is improved, and the burst loss is reduced.

Description

Technical Field

[0001] This invention relates to the field of fault detection technology, and in particular to a fault detection method and system for spinning machines based on machine learning. Background Technology

[0002] Spinning machines are the core equipment in chemical fiber production, and their operational stability directly determines filament quality and production efficiency. As the chemical fiber industry develops towards high capacity and precision, spinning machine structures are becoming increasingly complex, involving the coordinated operation of multiple systems such as melt conveying, winding, and heating temperature control. The causes of failures are multi-source and interconnected, making traditional fault detection methods insufficient to meet production demands. There are three main key shortcomings: Existing technologies mostly rely on manual inspection or the threshold alarm function built into the equipment. Manual inspection is limited by the cycle and it is difficult to capture instantaneous faults. Threshold alarms can only be triggered after the parameters exceed the safe range. By this time, the fault has already occurred, which can easily lead to the scrapping of a batch of filaments. It is impossible to achieve early warning and the timeliness of fault detection is seriously insufficient.

[0003] There is a strong nonlinear correlation between spinning machine faults and characteristic parameters (such as melt pressure and motor vibration). For example, the same pressure increase may correspond to spinneret blockage or screw wear. Traditional linear models are difficult to fit the complex relationship. At the same time, fault samples are scarce, and the model is prone to bias towards the majority class, resulting in a high rate of missed fault detection and false alarm, which cannot provide a reliable basis for operation and maintenance decisions.

[0004] Existing technology can only determine whether there is a fault, but cannot accurately locate the specific faulty component (e.g., knowing only that the melt pressure is abnormal, but not being able to distinguish whether it is a blocked spinneret, a leaking melt pipeline, or a failed metering pump). It requires maintenance personnel to check each component one by one based on their experience, with an average location time of more than 2 hours, which prolongs downtime and further aggravates production losses.

[0005] In summary, there is an urgent need for a machine learning-based method and system for detecting faults in spinning machines to meet the high-quality production needs of the chemical fiber industry. Summary of the Invention

[0006] This invention provides a machine learning-based method and system for fault detection in spinning machines, which addresses the shortcomings of existing technologies such as untimely fault detection, poor accuracy of prediction models, and inaccurate fault source localization.

[0007] On the one hand, the present invention provides a machine learning-based fault detection method for spinning machines, including: During the operation of the spinning machine, a fault detection interval is set, and process parameter data and result parameter data are collected after each fault detection interval.

[0008] The process parameter data and result parameter data are preprocessed to obtain comprehensive processed data, and fault-sensitive features are extracted from them.

[0009] A support vector machine (SVM) model is constructed, and its hyperparameters are optimized using the Genghis Khan Shark Optimization Algorithm to obtain a reinforced vector machine (SVM) model.

[0010] The fault-sensitive features are input into the reinforcement vector machine model, the fault prediction probability is output, and it is determined whether the preset fault judgment threshold is reached. If so, the fault source is located based on the current fault-sensitive features.

[0011] Develop a maintenance plan based on the source of the fault, and perform regular maintenance on the spinning machine according to the fault detection interval.

[0012] This invention provides a machine learning-based method for fault detection in spinning machines. Process parameters include melt temperature, melt pressure, winding speed, winding tension, motor current, motor vibration frequency, cooling air temperature and speed, and oil supply pressure and flow rate. Result parameters include filament diameter deviation, filament breaking strength, and filament evenness.

[0013] The present invention provides a machine learning-based method for fault detection in spinning machines, and the steps for obtaining comprehensively processed data include: For missing values ​​in process parameter data and result parameter data, corresponding strategies are used to supplement them based on the duration of the missing values.

[0014] For high-frequency noise generated by electromagnetic interference and mechanical vibration, noise reduction methods are selected according to different parameter characteristics to eliminate the noise.

[0015] For parameters with clear physical meaning and approximately normal distribution, Z-score standardization is used, while for parameters with fixed value range, Min-Max normalization is used.

[0016] The 3σ criterion is used to identify abnormal values ​​caused by sensor malfunctions and transient interference, and corresponding processing is carried out based on the number of consecutive abnormal points.

[0017] This invention provides a machine learning-based fault detection method for spinning machines, the steps of which include extracting fault-sensitive features: For faults involving aging of the heating rod and slow changes in screw wear, statistical features are generated by calculating basic statistics and trend characteristics.

[0018] For spinneret blockage and sudden tension breakage and change faults, the fluctuation characteristics and extreme value characteristics of the dynamic changes in the reaction are extracted as time-domain features.

[0019] For periodic faults such as motor bearing wear and gear meshing, the power spectral density is obtained through Fourier transform, and the characteristic frequencies whose energy proportion reaches a preset threshold are extracted as frequency domain features.

[0020] Calculate the Pearson correlation coefficients between statistical features, time-domain features, and frequency-domain features and fault labels, and use features that reach the preset correlation coefficient threshold as fault-sensitive features.

[0021] This invention provides a machine learning-based method for fault detection in spinning machines. The steps for constructing a support vector machine model include: The fault-sensitive features are divided into training and testing sets according to a preset ratio, and then class balancing is performed.

[0022] To address the nonlinear relationship between fault characteristics and fault types in spinning machines, a radial basis kernel function is selected.

[0023] We use grid search combined with cross-validation to optimize the penalty parameter and kernel function parameter as hyperparameters.

[0024] The model is trained using the support vector machine algorithm based on the training set and hyperparameters, and then validated using the test set to obtain the support vector machine model.

[0025] This invention provides a machine learning-based method for fault detection in spinning machines. The steps for obtaining the reinforcement vector machine model include: The initialization settings include the size of the shark population, with each individual shark representing a set of hyperparameters, and the fitness value of the support vector machine model is calculated using K-fold cross-validation.

[0026] During initialization and in each iteration, quasi-opposite solutions for the current shark individual are generated.

[0027] The fitness value of the quasi-opposite solution is calculated, and it is determined whether it is less than that of the current shark individual. If so, the current shark individual is replaced.

[0028] The position update formula of the Genghis Khan shark optimization algorithm is used to update the position of the individual shark, and the position of the individual shark is updated by combining the historical best position of the individual shark with the global best position of the individual shark.

[0029] The positions of individual sharks are sorted according to their fitness values. The neighborhoods of sharks that meet the preset requirements in the hyperparameter space are selected, and new solutions are generated using orthogonal arrays. The fitness values ​​are then calculated.

[0030] After reaching the preset number of iterations, the hyperparameters corresponding to the shark individual with the smallest fitness value are selected as the optimal hyperparameters, and a reinforcement vector machine model is constructed.

[0031] This invention provides a machine learning-based fault detection method for spinning machines, the steps of which include outputting fault prediction probabilities: The fault-sensitive features are transformed into a format that conforms to the input of the reinforcement vector machine model, and the decision values ​​are calculated.

[0032] The decision values ​​are converted into fault prediction probabilities by using Platt scaling to adapt to the probability quantification requirements of fault detection in spinning machines.

[0033] This invention provides a machine learning-based fault detection method for spinning machines, the steps of which include: From the current fault-sensitive features, select a preset number of fault-sensitive features that contribute the most to fault prediction as core features, and calculate the anomaly deviation.

[0034] A rule base for association is established based on the mechanical structure and process logic of the spinning machine, and matched with core features. The matching degree is calculated by combining the abnormal deviation degree with weighted summation.

[0035] Based on the matching degree and the association rule base, the initial fault source is located, and the fault range is narrowed down by combining the physical structure of the spinning machine and the sensor installation location to obtain the fault source.

[0036] This invention provides a machine learning-based method for detecting faults in spinning machines, and the steps for developing a maintenance plan include: The hazards of the failure source are assessed from three dimensions: production loss, equipment loss, and safety risk, and the maintenance time is determined in conjunction with the production plan.

[0037] The maintenance strategy is determined based on the type of fault source and the extent of damage, and preventive measures are developed to identify safety and equipment risks in operation.

[0038] On the other hand, the present invention provides a machine learning-based fault detection system for spinning machines, comprising: The multi-data acquisition module is used to set fault detection intervals during the operation of the spinning machine, and to collect process parameter data and result parameter data after each fault detection interval.

[0039] The fault feature extraction module is used to preprocess process parameter data and result parameter data to obtain comprehensive processed data, and extract fault-sensitive features from it.

[0040] The model optimization module is used to construct the support vector machine model. It optimizes the hyperparameters of the support vector machine model according to the Genghis Khan Shark Optimization Algorithm to obtain the reinforcement vector machine model.

[0041] The fault source determination module is used to input fault-sensitive features into the reinforcement vector machine model, output the fault prediction probability, and determine whether the preset fault judgment threshold is reached. If so, the fault source is located based on the current fault-sensitive features.

[0042] The solution development module is used to develop maintenance plans based on the source of the fault and to perform regular maintenance on the spinning machine according to the fault detection interval.

[0043] This invention provides a machine learning-based fault detection method and system for spinning machines. Through regular data collection and a series of processing and analysis steps, it can promptly detect faults and accurately determine their types, reducing production interruptions and product quality degradation caused by faults, and improving the timeliness and accuracy of fault detection. The data preprocessing method effectively addresses issues such as missing data, noise, and anomalies, providing a high-quality data foundation for subsequent fault detection and model training, thus improving the overall system reliability. The Genghis Khan Shark Optimization Algorithm efficiently optimizes the hyperparameters of the Support Vector Machine model, enabling the model to exhibit better performance in fault detection tasks and improving the accuracy and reliability of fault prediction. Furthermore, by combining multiple methods, it accurately locates the fault source, providing maintenance personnel with a clear direction for repairs, reducing maintenance time and costs, and improving equipment availability and production efficiency. It also comprehensively considers multiple dimensions to formulate maintenance plans, making maintenance more scientific and reasonable, ensuring normal equipment operation while minimizing the impact on production and reducing safety risks. Attached Figure Description

[0044] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0045] Figure 1 This is one of the flowcharts of the machine learning-based fault detection method for spinning machines provided in this embodiment of the invention; Figure 2 This is the second flowchart of the machine learning-based fault detection method for spinning machines provided in this embodiment of the invention. Figure 3 This is a flowchart illustrating the machine learning-based fault detection system for spinning machines provided in an embodiment of the present invention. Detailed Implementation

[0046] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0047] The following is combined with Figures 1-3 This invention describes a machine learning-based fault detection method and system for spinning machines.

[0048] like Figure 1 As shown, the machine learning-based fault detection method for spinning machines provided in this embodiment of the invention includes: During the operation of the spinning machine, a fault detection interval is set, and process parameter data and result parameter data are collected after each fault detection interval.

[0049] Process parameters include melt temperature, melt pressure, winding speed, winding tension, motor current, motor vibration frequency, cooling air temperature and speed, and oil supply pressure and flow rate. Result parameters include filament diameter deviation, filament breaking strength, and filament unevenness.

[0050] The process parameter data and result parameter data are preprocessed to obtain comprehensive processed data, and fault-sensitive features are extracted from them.

[0051] The steps to obtain comprehensive processed data include: For missing values ​​in process parameter data and result parameter data, corresponding strategies are used to supplement them based on the duration of the missing values.

[0052] If the missing duration is short (less than 5 sampling periods), linear interpolation is used: based on the changing trend of 3 valid data points before and after the missing point, the missing value is filled by linear fitting to ensure data continuity.

[0053] If the missing period is long (≥5 sampling periods), the KNN imputation method with similar operating conditions is adopted: select normal operation data segments with a similarity of >90% to the current operating conditions (such as raw material type, production specifications, and ambient temperature and humidity) from the historical database, and select the mean of the corresponding positions of the 5 most similar samples as the imputation value to avoid trend distortion caused by simple interpolation.

[0054] For high-frequency noise generated by electromagnetic interference and mechanical vibration, noise reduction methods are selected according to different parameter characteristics to eliminate the noise.

[0055] For slowly varying parameters such as melt temperature and melt pressure, a 5-point sliding window is used for smoothing: by calculating the average value of the current point and the two points before and after it, the impact of instantaneous fluctuations is reduced.

[0056] For transient parameters such as motor vibration frequency and winding tension, wavelet transform is used for noise reduction: a db4 wavelet basis is selected for 3-level decomposition, and soft thresholding is applied to high-frequency coefficients (to remove high-frequency components caused by noise) before reconstructing the signal to retain fault-related transient features.

[0057] For parameters with clear physical meaning and approximately normal distribution, Z-score standardization is used, while for parameters with fixed value range, Min-Max normalization is used.

[0058] The 3σ criterion is used to identify abnormal values ​​caused by sensor malfunctions and transient interference, and corresponding processing is carried out based on the number of consecutive abnormal points.

[0059] If the anomaly is a single sampling point (such as a sudden change in sensor readings), replace it with the average of the data from the previous and next time points to avoid the influence of isolated noise.

[0060] If three or more consecutive sampling points are abnormal, they are marked as "suspected faulty data segments" and the original values ​​are retained for subsequent fault feature analysis (such anomalies may be related to actual faults).

[0061] The steps for extracting fault-sensitive features include: For faults involving aging of the heating rod and slow changes in screw wear, statistical features are generated by calculating basic statistics and trend characteristics.

[0062] For spinneret blockage and sudden tension breakage and change faults, the fluctuation characteristics and extreme value characteristics of the dynamic changes in the reaction are extracted as time-domain features.

[0063] Fluctuation characteristics: Calculate the number of zero-crossing points of the parameter within the sliding window (the number of times it crosses the normal threshold; for example, frequent pressure exceeding the upper limit may indicate a risk of blockage) and the standard deviation of the difference between adjacent sampling points (reflecting the degree of parameter fluctuation).

[0064] Extreme value characteristics: Calculate the peak factor (the ratio of the peak value to the effective value within the window; the peak value of the parameter will increase significantly when there is a fault, such as when the spinneret is blocked, the pressure peak factor can reach more than 3 times the normal state) and the pulse factor (the ratio of the peak value to the mean value, used to amplify instantaneous abnormal signals).

[0065] For periodic faults such as motor bearing wear and gear meshing, the power spectral density is obtained through Fourier transform, and the characteristic frequencies whose energy proportion reaches a preset threshold are extracted as frequency domain features.

[0066] Calculate the Pearson correlation coefficients between statistical features, time-domain features, and frequency-domain features and fault labels, and use features that reach the preset correlation coefficient threshold as fault-sensitive features.

[0067] A support vector machine (SVM) model is constructed, and its hyperparameters are optimized using the Genghis Khan Shark Optimization Algorithm to obtain a reinforced vector machine (SVM) model.

[0068] The steps to build a support vector machine model include: The fault-sensitive features are divided into training and testing sets according to a preset ratio, and then class balancing is performed.

[0069] To address the nonlinear relationship between fault characteristics and fault types in spinning machines, a radial basis function kernel is selected, expressed by the following formula:

[0070] In the formula, These are kernel function parameters. This is the current sample. These are samples from the training set. It is the degree of similarity between two fault-sensitive features.

[0071] We use grid search combined with cross-validation to optimize the penalty parameter and kernel function parameter as hyperparameters.

[0072] The model is trained using the support vector machine algorithm based on the training set and hyperparameters, and then validated using the test set to obtain the support vector machine model.

[0073] like Figure 2 As shown, the steps to obtain the reinforcement vector machine model include: The initial shark population size is set, with each individual shark representing a set of hyperparameters. K-fold cross-validation is used to calculate the fitness value of the support vector machine model, expressed by the formula:

[0074] In the formula, It is the first The accuracy of cross-validation. It's a fold. It is the fitness value.

[0075] The quasi-complementary solution for the current shark individual is generated during initialization and in each iteration, expressed by the formula:

[0076] In the formula, It is the lower bound of the search space. It is the upper limit of the search space. It is a quasi-opposite solution.

[0077] The fitness value of the quasi-opposite solution is calculated, and it is determined whether it is less than that of the current shark individual. If so, the current shark individual is replaced.

[0078] The position update formula of the Genghis Khan shark optimization algorithm is used to update the position of a shark individual, combining the individual shark's historical best position and the global best individual shark position. The formula is expressed as follows:

[0079] In the formula, It is the first The first individual shark in the... The position of the next iteration. , It is a learning factor. It is a random number within the range [0,1]. It is the first The best historical position of an individual shark. It is the globally optimal position for the individual shark. This is the updated position of the individual shark.

[0080] The positions of individual sharks are sorted according to their fitness values. The neighborhoods of sharks that meet the preset requirements in the hyperparameter space are selected, and new solutions are generated using orthogonal arrays. The fitness values ​​are then calculated.

[0081] After reaching the preset number of iterations, the hyperparameters corresponding to the shark individual with the smallest fitness value are selected as the optimal hyperparameters, and a reinforcement vector machine model is constructed.

[0082] The fault-sensitive features are input into the reinforcement vector machine model, the fault prediction probability is output, and it is determined whether the preset fault judgment threshold is reached. If so, the fault source is located based on the current fault-sensitive features.

[0083] The steps for outputting the fault prediction probability include: The fault-sensitive features are transformed into a format consistent with the input of the reinforcement vector machine model, and the decision values ​​are calculated. The formula is expressed as follows:

[0084] In the formula, It's weight. It's a category label. It is the number of support vectors. It is a bias term. It is the decision value. Is the current test sample and the first Feature similarity of support vectors.

[0085] Platt scaling is used to convert the decision values ​​into fault prediction probabilities that meet the quantification requirements for fault detection in spinning machines. The formula is expressed as follows:

[0086] In the formula, , These are parameters that ensure the output satisfies the probability distribution characteristics. It represents the probability that the sample to be tested belongs to the fault category.

[0087] The probability of fault prediction can include probability values ​​for each category (e.g., "normal: 0.02, spinneret blockage: 0.97, motor failure: 0.01").

[0088] The most probable fault type (e.g., "spinneret blockage").

[0089] The steps to locate the source of the fault include: From the current fault-sensitive features, a predetermined number of fault-sensitive features that contribute the most to fault prediction are selected as core features, and the anomaly deviation is calculated. The formula is as follows:

[0090] In the formula, It is an abnormal deviation. It is the current eigenvalue. This is the average value under normal operating conditions. This is the standard deviation under normal operating conditions. For example, if the fault type is "spinneret blockage", the core characteristics are "melt pressure peak factor (contribution 45%)", "spinneret area temperature standard deviation (contribution 30%)", and "filament diameter deviation (contribution 18%)".

[0091] A rule base for association is established based on the mechanical structure and process logic of the spinning machine, and matched with core features. The matching degree is calculated by combining the anomaly deviation degree with a weighted summation, and the formula is expressed as:

[0092] In the formula, It's the degree of matching. It is the number of core features corresponding to the fault source in the association rule base. It is the feature weight. This is the current core feature.

[0093] Spinneret blockage: The core sensitive characteristics are a significant increase in the peak melt pressure factor, an increase in the standard deviation of temperature in the spinneret area, and an increase in filament diameter deviation. The process logic is that when the spinneret channels are blocked, the melt flow is obstructed, leading to a sudden increase in pressure. At the same time, the uneven distribution of melt temperature within the channels ultimately results in uneven filament formation and a larger diameter deviation.

[0094] Heating rod failure: The core sensitive characteristics are a decrease in the average temperature of the screw section, a decrease in the temperature slope (decreasing trend), and an increase in melt viscosity. The process logic is that the heating rod's heating capacity is insufficient, causing the screw section temperature to fail to meet the melt plasticization requirements, resulting in poor melt plasticization, which in turn increases the melt viscosity and affects the normal delivery of the melt.

[0095] Abnormal winding tension: The core sensitive characteristics are a significant increase in the slope of the winding tension sliding window, an increase in the frequency of motor current fluctuations, and an increase in the wire breakage rate. The process logic is that a malfunction in the winding tension control system leads to uneven stress on the wire. In order to maintain the winding state, the motor current fluctuates frequently, ultimately causing an increase in the wire breakage rate due to tension instability.

[0096] Motor bearing wear: The core sensitive characteristics are an increase in the motor vibration power spectral density (200-500Hz frequency band) and an increase in the vibration peak factor. The process logic is that after the motor bearing wears, the bearing clearance increases, generating periodic vibrations during operation, and forming characteristic vibration signals in the 200-500Hz frequency band, leading to abnormal vibration-related parameters.

[0097] Based on the matching degree and the association rule base, the initial fault source is located, and the fault range is narrowed down by combining the physical structure of the spinning machine and the sensor installation location to obtain the fault source.

[0098] The source of the fault is the faulty component and its location (e.g., "the spinneret's second to fourth group of orifices are blocked").

[0099] Develop a maintenance plan based on the source of the fault, and perform regular maintenance on the spinning machine according to the fault detection interval.

[0100] The steps to develop a maintenance plan include: The hazards of the failure source are assessed from three dimensions: production loss, equipment loss, and safety risk, and the maintenance time is determined in conjunction with the production plan.

[0101] Production loss: Calculate the direct losses caused by the failure (such as the amount of scrapped filaments due to spinneret blockage, and the production loss corresponding to the downtime).

[0102] Equipment damage: Determine whether the malfunction will lead to a chain of damage (e.g., if the heating rod fails and is not dealt with in time, the screw may become stuck due to poor plasticization of the melt).

[0103] Safety risks: Investigate whether the fault poses a safety hazard (such as high temperature caused by wear of motor bearings, which may lead to spontaneous combustion of lubricating oil).

[0104] Emergency malfunction (such as leakage in the molten metal pipeline): Immediately shut down the machine for maintenance, prioritizing equipment safety.

[0105] For minor faults (such as slight abnormalities in winding tension): schedule maintenance after the current production batch is completed to minimize production loss.

[0106] Chronic faults (such as a slow increase in motor vibration): Develop a maintenance plan within 72 hours and monitor the fault development trend simultaneously.

[0107] The maintenance strategy is determined based on the type of fault source and the extent of damage, and preventive measures are developed to identify safety and equipment risks in operation.

[0108] Maintenance strategies include addressing mechanical component failure sources (such as spinnerets, motor bearings, etc.): If the damage is minor (such as partial blockage of the spinneret orifices or slight abnormal noise from the motor bearing), the "cleaning / repair" strategy is applicable. For example, blockage of the spinneret orifices can be removed by flushing with hot melt, and slight abnormal noise from the motor bearing can be repaired by adding high-temperature grease.

[0109] If the damage is moderate or severe (such as spinneret channel deformation or motor bearing jamming), the "replacement" strategy applies. For example, if the spinneret channel wear exceeds the standard, a new spinneret needs to be replaced; if the motor bearing is jammed, the bearing of the same model needs to be replaced.

[0110] Electrical component failure sources (such as heating rods, pressure sensors, etc.): If the damage manifests as parameter drift (such as temperature control deviation of the heating rod or deviation of pressure sensor measurement value), a "calibration" strategy is applicable. For example, temperature control deviation of the heating rod can be corrected by recalibrating the temperature control module, and pressure sensor drift can be restored to accuracy by zero-point calibration.

[0111] If the damage manifests as functional failure (such as the heating rod not heating or the sensor not outputting a signal), the "replacement" strategy applies. For example, if the heating rod is burned out, it needs to be replaced with a heating rod of the same power; if the sensor is disconnected, it needs to be replaced with a new sensor.

[0112] Process-related fault sources (such as abnormal melt viscosity, not physical damage to components): If the damage manifests as parameters deviating from the normal range, a combination strategy of "process parameter adjustment + cleaning" should be applied. For example, if the viscosity is abnormal due to residual carbonization in the melt pipeline, the carbonized residue in the pipeline can be removed by high-temperature cleaning, and the screw speed or heating temperature can be adjusted to restore the melt viscosity to normal.

[0113] like Figure 3 As shown, based on the same general inventive concept, this invention also protects a machine learning-based fault detection system for spinning machines, the fault detection system comprising: The multi-data acquisition module is used to set fault detection intervals during the operation of the spinning machine, and to collect process parameter data and result parameter data after each fault detection interval.

[0114] The fault feature extraction module is used to preprocess process parameter data and result parameter data to obtain comprehensive processed data, and extract fault-sensitive features from it.

[0115] The model optimization module is used to construct the support vector machine model. It optimizes the hyperparameters of the support vector machine model according to the Genghis Khan Shark Optimization Algorithm to obtain the reinforcement vector machine model.

[0116] The fault source determination module is used to input fault-sensitive features into the reinforcement vector machine model, output the fault prediction probability, and determine whether the preset fault judgment threshold is reached. If so, the fault source is located based on the current fault-sensitive features.

[0117] The solution development module is used to develop maintenance plans based on the source of the fault and to perform regular maintenance on the spinning machine according to the fault detection interval.

[0118] The machine learning-based fault detection method and system provided in this embodiment significantly improves the sensitivity and accuracy of fault detection by employing multi-dimensional feature extraction methods for different fault types and filtering features strongly correlated with fault labels using Pearson correlation coefficients. Furthermore, it optimizes SVM hyperparameters using the Genghis Khan Shark Optimization Algorithm, expands the search space using quasi-opposites learning, and refines the local search using orthogonal learning. This effectively avoids the high computational cost of traditional grid search and the tendency of random search to get trapped in local optima, thus improving the model's generalization ability. It also quickly locates the fault source by calculating the abnormal deviation of fault-sensitive features and the contribution of core features, combined with a rule base relating the spinning machine's mechanical structure and process logic. Simultaneously, it assesses the fault's impact from three dimensions: production loss, equipment loss, and safety risk, and formulates targeted maintenance plans, reducing maintenance costs and downtime.

[0119] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0120] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A machine learning based spinning machine fault detection method, characterized by, The application relates to a method for predicting spinning machine faults. During the operation of the spinning machine, a fault detection interval is set, and process parameter data and result parameter data are collected every time the fault detection interval is experienced; The process parameter data and the result parameter data are preprocessed to obtain comprehensive processing data, and fault sensitive features are extracted therefrom; A support vector machine model is constructed, and hyperparameters of the support vector machine model are optimized according to a Gengiskhan shark optimization algorithm to obtain a reinforced support vector machine model; The fault sensitive features are input into the reinforced support vector machine model, a fault prediction probability is output, and it is determined whether a preset fault judgment threshold is reached. If yes, a fault source is located according to the current fault sensitive features; A maintenance scheme is formulated according to the fault source, and the spinning machine is regularly maintained according to the fault detection interval.

2. The machine learning based spin machine fault detection method of claim 1, wherein, The process parameter data include melt temperature, melt pressure, winding speed, winding tension, motor current, motor vibration frequency, cooling air temperature and air speed, oil supply pressure and flow; and the result parameter data include yarn diameter deviation, yarn breaking strength and yarn evenness. 3.The machine learning based spindle failure detection method of claim 1, wherein, The step of obtaining the comprehensive processing data comprises the following steps: For missing values in the process parameter data and the result parameter data, corresponding strategies are adopted for supplement according to missing time lengths; For high-frequency noise generated by electromagnetic interference and mechanical vibration, different parameter characteristics are selected to select a denoising method to eliminate the noise; For parameters with clear physical meaning and approximately normal distribution, Z-score standardization is adopted; and for parameters with fixed value range, Min-Max normalization is adopted; Through a 3sigma criterion, abnormal values caused by sensor faults and transient interference are identified, and corresponding processing is performed according to the number of continuous abnormal points.

4. The machine learning based spin machine fault detection method of claim 3, wherein, The step of extracting the fault sensitive features comprises the following steps: For slowly changing faults of a heating rod and screw wear, statistical features are generated by calculating basic statistics and trend features; For faults of spinneret blockage and sudden breaking of tension, fluctuation features and extreme value features that dynamically change are extracted as time domain features; For faults of motor bearing wear and gear meshing with periodicity, power spectral density is obtained through Fourier transform, and feature frequencies with energy proportion reaching a preset threshold are extracted as frequency domain features; Pearson correlation coefficients of the statistical features, the time domain features and the frequency domain features and fault labels are calculated, and features reaching a preset correlation coefficient threshold are taken as the fault sensitive features. 5.The machine learning based spindle failure detection method of claim 1, wherein, The step of constructing the support vector machine model comprises the following steps: The fault sensitive features are divided into a training set and a test set according to a preset proportion, and class balancing processing is performed; For nonlinear relationships between spinning machine fault features and fault types, a radial basis kernel function is selected; Grid search combined with cross-validation is adopted to optimize penalty parameters and kernel function parameters as the hyperparameters; Based on the training set and the hyperparameters, a support vector machine algorithm is used for model training, and the support vector machine model is obtained through verification according to the test set. 6.The machine learning based spindle failure detection method of claim 1, wherein, The step of obtaining the reinforced support vector machine model comprises the following steps: An initial shark population size is set, each shark individual represents a group of hyperparameters, and the fitness value of the support vector machine model is calculated through K-fold cross-validation; Generating a quasi-antithetic solution of the current shark individual in initialization and each iteration; And calculating the fitness value of the quasi-antithetic solution, judging whether it is less than the current shark individual, if so, replacing the current shark individual; According to the position updating formula of the Genghis Khan shark optimization algorithm, and combining the historical optimal position of the shark individual and the position of the global optimal shark individual, the position of the shark individual is updated; According to the size of the fitness value, the positions of the shark individuals are sorted, the neighborhood of the shark individual reaching the preset requirement in the hyperparameter space is selected, a new solution is generated using an orthogonal table, and the fitness value is calculated; After reaching the preset number of iterations, the hyperparameters corresponding to the shark individual with the smallest fitness value are selected as the optimal hyperparameters, and the reinforced vector machine model is constructed. 7.The machine learning based spindle failure detection method of claim 1, wherein, The step of outputting the failure prediction probability comprises: Convert the failure sensitive features into a format that meets the input of the reinforced vector machine model, and calculate the decision value; Through Platt scaling, the decision value is converted into the requirement of quantifying the possibility of failure detection of the spinning machine as the failure prediction probability. 8.The machine learning based spindle failure detection method of claim 1, wherein, The step of locating the failure source comprises: From the current failure sensitive features, a preset number of failure sensitive features with the highest contribution to failure prediction are selected as core features, and the abnormal deviation degree is calculated; Based on the mechanical structure and process logic of the spinning machine, an association rule base is established, and is matched with the core features, and the matching degree is calculated through the abnormal deviation degree and combined with weighted summation; According to the matching degree and the association rule base, the preliminary failure source is located, and the failure range is narrowed to obtain the failure source combined with the physical structure of the spinning machine and the installation position of the sensor. 9.The machine learning based spindle failure detection method of claim 1, wherein, The step of formulating the maintenance scheme comprises: From the production loss, equipment loss and safety risk, the harm of the failure source is evaluated, and the maintenance time is determined combined with the production plan; According to the type and damage degree of the failure source, the maintenance strategy is determined, and the safety risk and equipment risk in operation are identified to formulate prevention and control measures.

10. A machine learning based spinning machine fault detection system applied to a machine learning based spinning machine fault detection method according to any one of claims 1 to 9, characterized in that, The failure detection system comprises: A multi-data acquisition module is used to set a failure detection interval during the operation of the spinning machine, and process parameter data and result parameter data are collected every failure detection interval; A failure feature extraction module is used to preprocess the process parameter data and the result parameter data to obtain comprehensive processing data, and extract failure sensitive features therefrom; A model optimization module is used to construct a support vector machine model, and optimize the hyperparameters of the support vector machine model according to the Genghis Khan shark optimization algorithm, to obtain a reinforced vector machine model; A failure source determination module is used to input the failure sensitive features into the reinforced vector machine model, output a failure prediction probability, and judge whether a preset failure judgment threshold is reached, if so, locate the failure source according to the current failure sensitive features; A scheme formulation module is used to formulate a maintenance scheme according to the failure source, and periodically maintain the spinning machine according to the failure detection interval.

Images