A hot stamping equipment fault detection system and method based on data acquisition and analysis

Abstract

The present invention discloses a hot stamping equipment fault detection system and method based on data collection and analysis, which relates to the technical field of equipment fault detection. The following steps are as follows: first, various operating parameter information generated during the operation of the hot stamping equipment is obtained. At the same time, a data storage and management system is established to effectively save and organize the large amount of collected data to facilitate subsequent calls and processing. The present invention regularly updates the model by setting a standard model training frequency to ensure that it adapts to the state changes and known faults of the hot stamping equipment, maintains high detection accuracy, reduces false alarms and missed alarms, and ensures the stability of the production line. For unrecognizable faults, the solution dynamically adjusts the training frequency so that the model can quickly learn new fault characteristics, improve the recognition rate, and reduce the risk of downtime and equipment damage due to the failure to detect new faults in a timely manner, thereby improving production efficiency and product quality reliability.

Description

Technical Field

[0001] The present invention relates to the technical field of equipment fault detection, and in particular to a hot stamping equipment fault detection system and method based on data collection and analysis. Background Art

[0002] Hot stamping equipment fault detection based on data collection and analysis involves collecting various data (such as temperature, pressure, and speed) in real time during hot stamping equipment operation and using data analysis techniques to identify anomalies or potential faults in the equipment's operating status. Data is first collected through sensors or other monitoring devices. This data is then preprocessed and feature extracted to analyze for abnormal signals. Finally, based on the analysis results, a determination is made as to whether the equipment is faulty or requires maintenance. This approach can help prevent equipment downtime and quality issues, improving production efficiency and equipment lifespan.

[0003] In hot stamping equipment fault detection based on data collection and analysis, feedback optimization and model updates refer to feeding new data, fault information, and test results collected during actual equipment operation back into the system to retrain or adjust the fault detection model to ensure its continued accuracy. Feedback optimization helps the system continuously adjust the detection model based on the equipment's operating conditions and environmental changes, enhancing its ability to identify new fault modes. Model updates ensure that the system maintains a high level of fault detection effectiveness throughout the equipment's lifecycle, enabling real-time monitoring and accurate early warning, thereby extending equipment lifespan and production efficiency.

[0004] The existing technology has the following deficiencies:

[0005] Existing technologies typically periodically add collected equipment operation data to training datasets, retraining and optimizing the models to maintain detection accuracy, which meets requirements in most scenarios. However, when hot stamping equipment frequently experiences new failure modes or types, regular model training alone may not be able to capture these new features in a timely manner, resulting in delayed detection results. If the model fails to quickly adapt to new failure modes, the detection system will struggle to identify and warn of failures in a timely manner, causing undetected failures to gradually accumulate during production, leading to frequent equipment failures and downtime, significantly reducing production efficiency. This cumulative effect can also have a chain reaction on other related equipment, causing equipment damage and product quality issues, ultimately endangering the stability of the entire production line and even causing production to stop.

[0006] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not form the prior art that is already known to a person of ordinary skill in the art. Summary of the Invention

[0007] The purpose of the present invention is to provide a hot stamping equipment fault detection system and method based on data collection and analysis. By setting a standard model training frequency and regularly updating the model, the system ensures that it adapts to the state changes and known faults of the hot stamping equipment, maintains high detection accuracy, reduces false positives and missed negatives, and ensures production line stability. For unrecognizable faults, the system dynamically adjusts the training frequency, allowing the model to quickly learn new fault characteristics, improve recognition rate, and reduce the risk of downtime and equipment damage caused by untimely detection of new faults, thereby improving production efficiency and product quality reliability, and addressing the problems mentioned above.

[0008] In order to achieve the above object, the present invention provides the following technical solution: a method for detecting a fault of a hot stamping device based on data collection and analysis, comprising the following steps:

[0009] First, obtain the various operating parameter information generated during the operation of the hot stamping equipment. At the same time, establish a data storage and management system to effectively save and organize the large amount of collected data to facilitate subsequent call and processing;

[0010] Perform feature extraction on pre-processed hot stamping equipment operating parameter data. Based on the extracted fault feature information, a pre-learned machine learning model is used to intelligently evaluate the changes in new fault modes generated during the operation of the hot stamping equipment.

[0011] Based on the results of intelligent evaluation of the machine learning model, the fault types during the operation of the hot stamping equipment are divided into identifiable fault types and unidentifiable fault types;

[0012] Set a standard model training frequency for identifiable fault types. Regularly train the model to ensure it can adapt to various changes in the equipment, maintain high detection accuracy for known faults, and improve detection accuracy and applicability.

[0013] For unrecognizable fault types, the model training frequency is dynamically adjusted based on the standard model training frequency set for recognizable fault types, so that the model can capture the characteristics of new faults in a timely manner and improve the recognition rate.

[0014] Preferably, the steps for obtaining various operating parameter information generated during the operation of the hot stamping equipment and establishing a data storage and management system to effectively save and organize the large amount of collected data for subsequent retrieval and processing are as follows:

[0015] After obtaining various parameters during the operation of the hot stamping equipment, data collection and real-time transmission are first performed;

[0016] Clean and format the collected data to ensure data quality and consistency;

[0017] Store cleaned and formatted data in a database and establish an efficient storage and indexing system to ensure that data can be quickly retrieved and managed;

[0018] Establish a complete data management system and set access permissions to ensure that data is effectively protected while being used efficiently.

[0019] Preferably, feature extraction is performed on the pre-processed hot stamping equipment operating parameter data, and the extracted features include the complexity of the signal and the symmetry of the data distribution. After acquisition, the complexity of the signal and the symmetry of the data distribution are analyzed under the detection window to generate a complexity index and a skewness index respectively. The complexity index quantifies the complexity of the signal, that is, the diversity and instability of the signal changes in the time or frequency domain, reflecting the fluctuation amplitude and uncertainty of the equipment operation status. The skewness index quantifies the symmetry of the data distribution, that is, whether the data is symmetrically distributed around the average value, reflecting the degree of offset and abnormal deviation trend in the equipment status data.

[0020] Preferably, after obtaining the complexity index and skewness index, the generated complexity index and skewness index are input into a pre-learned machine learning model, a fault identification coefficient is generated by the machine learning model, and the fault identification coefficient is used to perform an intelligent evaluation of the newly generated fault mode changes during the operation of the hot stamping equipment.

[0021] Preferably, the fault identification coefficient generated after comprehensive analysis of the complexity index and the skewness index is compared with a preset fault identification coefficient reference threshold value to classify the fault types during the operation of the hot stamping equipment. The classification steps are as follows;

[0022] If the fault identification coefficient is greater than or equal to a preset fault identification coefficient reference threshold, the fault type during the operation of the hot stamping equipment is classified as an unrecognizable fault type;

[0023] If the fault identification coefficient is less than a preset fault identification coefficient reference threshold, the fault type during the operation of the hot stamping equipment is classified as an identifiable fault type.

[0024] Preferably, for unrecognizable fault types, the model training frequency is dynamically adjusted based on the standard model training frequency set for recognizable fault types, so that the model can capture the characteristics of new faults in a timely manner and improve the recognition rate. The specific steps are as follows:

[0025] The standard model training frequency based on the identifiable fault type is calibrated as F std During operation, the fault identification coefficient FIC is used to quantify the faults detected in the current state of the equipment. The fault identification coefficient reference threshold FIC refIt is used to quantify the model's recognition criteria for known faults. If the fault recognition coefficient FIC is greater than or equal to the fault recognition coefficient reference threshold FIC ref , it means that there are new, unrecognizable fault types, and the model needs to be dynamically adjusted to improve the ability to identify new faults;

[0026] During detection, the fault identification coefficient FIC is compared with the fault identification coefficient reference threshold FIC ref For comparison:

[0027] If FIC≥FIC ref , then an unrecognizable fault type has occurred, and the model training frequency needs to be dynamically adjusted according to the severity of the fault;

[0028] If FIC <FIC ref , indicating that the fault type can be identified by the current model, there is no need to change the training frequency, and the standard frequency F std ;

[0029] When FIC≥FIC ref , according to the fault identification coefficient FIC for the fault identification coefficient reference threshold FIC ref The training frequency is dynamically adjusted based on the specific percentage of excess, and is calculated using the following formula;

[0030]

[0031] , where F new is the adjusted training frequency, and H is the adjustment coefficient, which is used to control the increase in training frequency;

[0032] Perform the following operations based on the magnitude by which the fault identification coefficient exceeds the reference threshold:

[0033] If FIC exceeds FIC ref If the percentage is between 0-30%, set H=2 and the training frequency F new Doubled to accommodate the current minor anomaly;

[0034] If FIC exceeds FIC ref If the percentage exceeds 30%, set H=4 and the training frequency F new Improved by three times to accommodate complex fault characteristics and ensure that the model can capture new failure modes in a timely manner;

[0035] With the frequent training of the model, the changes of the fault identification coefficient FIC are monitored in real time. If the fault identification coefficient FIC recovers to a value lower than the fault identification coefficient reference threshold FIC during continuous monitoring, ref , indicating that the model has learned new fault features and restored the training frequency to the standard frequency F std ;

[0036] if FIC <FIC ref ,then F new =F std

[0037] At this point, by restoring the training frequency to the standard frequency, we can maintain the model's efficiency and rational resource utilization, while ensuring the stability of fault detection after the model has adapted to the new features.

[0038] On the basis of dynamically adjusting the model training frequency, continue to monitor the changing trend of the equipment status and fault identification coefficient FIC. If unrecognizable fault types continue to appear frequently, reset the standard frequency F std , to adapt to long-term changes in equipment and improve the adaptability of fault detection models.

[0039] Preferably, within the detection window, the complexity of the signal is analyzed to generate a complexity index in the following specific steps:

[0040] In the detection window, the pre-processed hot stamping equipment operation signal is subjected to wavelet transform to decompose it into time-frequency features of different scales and frequency components. The calculation expression is as follows:

[0041]

[0042] , where W x (a, b) are the wavelet transform coefficients, i.e., the signal characteristics captured by the pre-processed hot stamping equipment operation signal x(t) at a given scale a and position b. Scale a is the scale factor, which is used to control the expansion and contraction of the mother wavelet. Position b is the translation factor, which is used to control the position of the mother wavelet on the time axis, indicating the local change of the signal during this period. * is the complex conjugate of the mother wavelet, ψ is the wavelet function used to decompose the signal;

[0043] For the wavelet transform coefficient W x (a, b) Calculate the sum of squares to calculate the energy distribution of the signal at each scale. The calculation expression is as follows:

[0044]

[0045] , where E(a) is the energy distribution, i.e. the energy distribution at scale a;

[0046] The sample entropy is calculated at each scale to quantify the randomness and complexity of the signal at that scale. The calculation expression is as follows:

[0047]

[0048] , where is the sample entropy, m is the embedding dimension, r is the similarity tolerance, N is the total number of samples, which defines the length of the entire signal and the analysis range, and i represents the index position of the sample, that is, the position of each data point in the signal. is the number of similar patterns of the i-th sample under the embedding dimension m and similarity tolerance r, is the number of similar patterns of the i-th sample under the embedding dimension m+1 and similarity tolerance r;

[0049] The normalized energy distribution of each scale is combined with the corresponding sample entropy to generate a complexity index to quantify the overall complexity of the signal. The calculation expression is as follows:

[0050]

[0051] , where Complexity Index is the complexity index, is the sample entropy on scale a.

[0052] Preferably, within the detection window, the symmetry of the data distribution is analyzed to generate the skewness index in the following specific steps:

[0053] In the detection window, the pre-processed hot stamping equipment operation signal is calibrated as x(t), and the difference between each sampling point and its adjacent points is calculated. The expression is as follows:

[0054] d z =|x(t z )-x(t z+1 )|

[0055] , where d z is the difference between adjacent points, x(t z ) is at the zth time point t z The signal value collected at the moment, x(t z+1 ) is at time t z A time point after time t z+1 The collected signal value;

[0056] The difference d between all adjacent points z , calculate its symmetry weight to determine the symmetry characteristics of the signal. The calculation expression is as follows:

[0057]

[0058] , where W s is the symmetry weight, Q is the total number of differences;

[0059] Based on the symmetry weight W s, calculate the weighted skewness value of the signal to quantify the symmetry of the data distribution. The calculation expression is as follows:

[0060]

[0061] , where K is the weighted skewness value;

[0062] Normalize the weighted skewness value K to generate the skewness index. The calculation expression is as follows:

[0063]

[0064] Where, Skewness Index is the skewness index, and max(|K|) is the maximum skewness value observed in the sample data set.

[0065] A hot stamping equipment fault detection system based on data collection and analysis includes a data collection and management module, a feature extraction and intelligent evaluation module, a fault type classification module, a regular update module for identifiable faults, and a dynamic adaptation module for unidentifiable faults:

[0066] The data collection and management module first obtains the various operating parameter information generated during the operation of the hot stamping equipment. At the same time, it establishes a data storage and management system to effectively save and organize the large amount of collected data for subsequent retrieval and processing.

[0067] The feature extraction and intelligent evaluation module extracts features from pre-processed hot stamping equipment operating parameter data. Based on the extracted fault feature information, it uses a pre-learned machine learning model to intelligently evaluate new fault mode changes generated during the hot stamping equipment operation.

[0068] The fault type classification module, based on the results of intelligent evaluation of the machine learning model, divides the fault types during the operation of the hot stamping equipment into identifiable fault types and unidentifiable fault types;

[0069] The identifiable fault regular update module sets a standard model training frequency for identifiable fault types. By regularly training the model, it ensures that it can adapt to various changes in the equipment, maintains high detection accuracy for known faults, and improves detection accuracy and applicability.

[0070] The unrecognizable fault dynamic adaptation module dynamically adjusts the model training frequency for unrecognizable fault types based on the standard model training frequency set for recognizable fault types, so that the model can capture the characteristics of new faults in a timely manner and improve the recognition rate.

[0071] In the above technical solution, the technical effects and advantages provided by the present invention are:

[0072] The present invention regularly updates the model by setting a standard model training frequency, thereby ensuring that the model can continuously adapt to the normal state changes and known fault modes of the hot stamping equipment. For identifiable fault types, regular model updates can help the model maintain high detection accuracy for these known faults and reduce the occurrence of false positives and missed positives. In a production environment, this strategy can quickly detect known fault modes and issue early warnings in a timely manner to prevent small faults from accumulating into large faults, helping to ensure the stability of the production line and the operating efficiency of the equipment. In addition, this solution establishes a data storage and management system to provide a reliable data foundation for subsequent analysis and feature extraction, allowing the entire fault detection system to be monitored and maintained in real time during equipment operation, reducing production risks caused by equipment downtime or abnormalities.

[0073] The present invention targets unrecognizable fault types, and the solution adopts a strategy of dynamically adjusting the frequency of model training so that the model can capture and learn the characteristics of new faults in a timely manner. When the fault recognition coefficient reaches or exceeds the reference threshold, the system will identify the fault as an unrecognizable fault, thereby adding new data to the training set through higher-frequency training, prompting the model to quickly adapt to new fault modes. This not only enables the timely identification of new fault modes, but also effectively improves the adaptability and recognition rate of the model. Through this strategy of dynamically adjusting the training frequency, the risk of production downtime and equipment damage caused by new faults not being detected in a timely manner can be significantly reduced, and problems can be prevented from accumulating during equipment operation to the point of endangering the stability of the production line or even causing production stoppages, thereby improving production efficiency and product quality reliability. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction to the drawings required for use in the embodiments will be given below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0075] Figure 1 The present invention is a method flow chart of a method for detecting faults of hot stamping equipment based on data collection and analysis.

[0076] Figure 2 The figure is a module diagram of a hot stamping equipment fault detection system based on data collection and analysis according to the present invention. DETAILED DESCRIPTION

[0077] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, these example embodiments are provided so that the description of this disclosure will be thorough and complete and will fully convey the concepts of the example embodiments to those skilled in the art.

[0078] The present invention provides Figure 1 A method for detecting faults in hot stamping equipment based on data collection and analysis is shown, comprising the following steps:

[0079] First, obtain the various operating parameter information generated during the operation of the hot stamping equipment. At the same time, establish a data storage and management system to effectively save and organize the large amount of collected data to facilitate subsequent call and processing;

[0080] The steps to obtain the various operating parameter information generated during the operation of the hot stamping equipment and establish a data storage and management system to effectively save and organize the large amount of collected data for subsequent call and processing are as follows:

[0081] After obtaining various parameters during the operation of the hot stamping equipment, data collection and real-time transmission are first performed;

[0082] The collected parameter information includes key operational data such as temperature, pressure, speed, and current during device operation. The data collection process utilizes a high-speed data transmission protocol to transmit sensor data in real time to a central processing unit or server. Real-time transmission ensures data timeliness and integrity, preventing data accuracy from being compromised by delays or loss. The frequency and interval of data collection can also be configured to ensure that the data comprehensively reflects device status while minimizing the transmission and storage burdens caused by excessive data volume.

[0083] Clean and format the collected data to ensure data quality and consistency;

[0084] Because real-time data may contain outliers, duplicates, or missing data due to transmission failures, it is necessary to clean the data using methods such as denoising, interpolation, and data completion. Furthermore, sensors from different sources may output data in different formats, requiring unified format conversion. For example, converting temperature to standard units and synchronizing timestamps to a unified format ensures data compatibility during subsequent processing. Cleaned and formatted data not only facilitates storage but also allows for direct use in subsequent analysis and feature extraction.

[0085] Store cleaned and formatted data in a database and establish an efficient storage and indexing system to ensure that data can be quickly retrieved and managed;

[0086] Common storage methods include relational databases, NoSQL databases, or time series databases, with the choice depending on the data type and usage requirements. For example, a time series database is suitable for storing time series information of device operating data. To facilitate subsequent access and processing, multi-level indexes should be established in the database, such as those based on time, device number, or data type. This makes querying and processing data for specific time periods, devices, or types more efficient. Furthermore, data storage should incorporate backup and redundancy mechanisms to ensure data security.

[0087] Establish a comprehensive data management system and set access permissions to ensure data is effectively protected while being used efficiently;

[0088] The data management system includes data classification, archiving, and lifecycle management, such as archiving historical data and assigning high priority to real-time data. Access rights are divided according to user identity and responsibilities to ensure data security and privacy and prevent unauthorized access. The management system should also include data retrieval, update, and deletion capabilities to support subsequent data analysis, fault detection model training, and report generation. The system should also ensure good scalability to accommodate future data growth.

[0089] Perform feature extraction on pre-processed hot stamping equipment operating parameter data. Based on the extracted feature information, a pre-learned machine learning model is used to intelligently evaluate the changes in new fault modes generated during the operation of the hot stamping equipment.

[0090] Feature extraction is performed on the pre-processed hot stamping equipment operating parameter data. The extracted features include the complexity of the signal and the symmetry of the data distribution. After acquisition, the signal complexity and the symmetry of the data distribution are analyzed within the detection window to generate the complexity index and skewness index respectively. The complexity index quantifies the complexity of the signal, that is, the diversity and instability of the signal changes in the time or frequency domain, reflecting the fluctuation amplitude and uncertainty of the equipment operation status. The skewness index quantifies the symmetry of the data distribution, that is, whether the data is symmetrically distributed around the mean value, reflecting the degree of deviation and abnormal deviation trend in the equipment status data.

[0091] When the complexity of the signal increases significantly, this may indicate that the acquired fault features are of a fault type that the current model cannot recognize, because the model is usually trained on known, recognizable fault feature patterns and has a certain range of adaptability to the complexity of these patterns. However, when new fault features appear, they often cause more diverse frequency components, randomness, or nonlinear changes in the signal, causing the complexity index to increase significantly. At this time, because the model has never seen similar complex features, it is difficult to accurately match or classify these new signals, resulting in a reduced recognition rate or even misjudgment. This sudden increase in complexity exceeds the model's learning experience, indicating that the current features may belong to a new fault type, and the model training needs to be dynamically adjusted to capture and learn the new fault features and achieve accurate identification and monitoring of the equipment status.

[0092] In the detection window, the complexity of the signal is analyzed and the specific steps to generate the complexity index are as follows:

[0093] In the detection window, the pre-processed hot stamping equipment operation signal is subjected to wavelet transform to decompose it into time-frequency features of different scales and frequency components. The calculation expression is as follows:

[0094]

[0095] , where W x (a, b) are the wavelet transform coefficients, i.e., the signal characteristics captured by the pre-processed hot stamping equipment operation signal x(t) at a given scale a and position b. Scale a is the scale factor, which is used to control the expansion and contraction of the mother wavelet. Position b is the translation factor, which is used to control the position of the mother wavelet on the time axis, indicating the local change of the signal during this period. * is the complex conjugate of the mother wavelet, ψ is the wavelet function used to decompose the signal. If the wavelet is real, then ψ * =ψ, in wavelet transform, complex conjugate is used to process complex-valued signals to ensure that the transform result contains complete information of the signal;

[0096] In the wavelet transform, a given scale represents the frequency level at which the signal is decomposed, which determines the degree of "stretching" of the wavelet: smaller scales correspond to high-frequency components (capturing rapid changes and details in the signal), while larger scales correspond to low-frequency components (capturing the overall trend and slow changes in the signal). A given position refers to the specific location where the wavelet moves on the time axis, controlling the alignment of the wavelet function in the signal. By adjusting the scale and position, the wavelet transform can analyze the local characteristics of the signal within different time and frequency ranges, capturing the transient characteristics or short-term anomalies of the signal, making detailed local analysis possible in the time-frequency space. This time-frequency localization capability makes the wavelet transform particularly suitable for analyzing non-stationary signals, such as monitoring and identifying faults in hot stamping equipment.

[0097] Through wavelet transform, the signal can be decomposed into a series of components of different scales, which is suitable for detecting the characteristics of the signal at different frequencies, especially the abnormal or complex changes in the device signal.

[0098] For the wavelet transform coefficient W x (a, b) Calculate the sum of squares to calculate the energy distribution of the signal at each scale. The calculation expression is as follows:

[0099]

[0100] Where E(a) is the energy distribution, that is, the energy distribution at scale a, which is used to measure the energy intensity of the signal at that scale;

[0101] Calculating the signal's energy distribution at each scale reveals the concentration of energy across different frequency components, helping to analyze the frequency bands where its key characteristics reside. For complex systems like hot stamping equipment, different fault types may exhibit distinct frequency characteristics. For example, high-frequency components may reflect vibration or rapidly changing faults, while low-frequency components may correspond to slower-moving system anomalies. By calculating the energy distribution at each scale, the signal's strength in each frequency band can be identified, highlighting key fault characteristic frequency bands. This allows the fault detection model to more accurately capture abnormal conditions, improving the accuracy and sensitivity of fault identification.

[0102] The sample entropy is calculated at each scale to quantify the randomness and complexity of the signal at that scale. The calculation expression is as follows:

[0103]

[0104] Where, is the sample entropy, which is used to quantify the complexity or uncertainty of the signal, m is the embedding dimension, r is the similarity tolerance, which represents the distance threshold at which two patterns are considered similar, N is the total number of samples, which defines the length of the entire signal and the scope of analysis, and i represents the index position of the sample, that is, the position of each data point in the signal. is the number of similar patterns of the i-th sample under the embedding dimension m and similarity tolerance r, is the number of similar patterns of the i-th sample under the embedding dimension m+1 and similarity tolerance r;

[0105] The total number of samples minus the embedding dimension (i.e., (Nm)) is used in the sample entropy calculation to ensure that each extracted subsequence (pattern) has sufficient length in the signal for comparison and matching.

[0106] Specifically, sample entropy is used to measure the similarity of subsequences in a signal, and the embedding dimension m determines the length of the subsequence. If the pattern is extracted starting from the last sample of the signal, the pattern may not reach the required embedding dimension length. Therefore, to avoid this, when calculating the sample entropy, only the positions from the 1st to the (Nm)th sample are considered as the starting point to ensure that each subsequence has a sufficient length m for comparison.

[0107] By using (Nm), the sample entropy formula can accurately compare the similarity of each valid subsequence in the signal, thereby more reliably quantifying the complexity and randomness of the signal.

[0108] The normalized energy distribution of each scale is combined with the corresponding sample entropy to generate a complexity index to quantify the overall complexity of the signal. The calculation expression is as follows:

[0109]

[0110] Where Complexity Index is the complexity index, is the sample entropy at scale a, quantifying the randomness and complexity at that scale;

[0111] Within the detection window, a larger value for the complexity index, generated after analyzing the signal's complexity, generally indicates greater signal complexity and uncertainty, which may mean that the acquired fault signature is a fault type unrecognizable by the current model. When the complexity index increases significantly within the monitoring window, it indicates that the pattern or frequency components in the signal have become complex, presenting new features that the model has not learned, resulting in the model's inability to accurately match or classify, thereby reducing recognition accuracy. Conversely, if the complexity index remains low, the signal pattern is generally within the model's known range, indicating that the acquired fault signature is likely a type recognizable by the current model, and the model can normally detect and classify the fault condition.

[0112] When the symmetry of the data distribution changes significantly, it usually indicates that the acquired fault features may belong to a fault type that the current model cannot recognize. This is because the machine learning model relies on the feature distribution patterns under known faults and normal conditions during training, and establishes associations between these patterns. If the symmetry of the data (such as the skewness index) deviates significantly from the symmetrical distribution expected by the model, it may mean that new fault modes or characteristics have emerged, and these characteristics are not fully represented in the training data. The model cannot associate such asymmetric features with known faults, resulting in reduced recognition accuracy. This significant distribution change indicates a potential new fault mode, suggesting that data needs to be re-collected and the model needs to be updated in order to adapt to and recognize such new fault features in a timely manner.

[0113] Under the detection window, the symmetry of the data distribution is analyzed and the specific steps to generate the skewness index are as follows:

[0114] In the detection window, the pre-processed hot stamping equipment operation signal is calibrated as x(t), and the difference between each sampling point and its adjacent points is calculated. The expression is as follows:

[0115] d z =|x(t z )-x(t z+1 )|

[0116] , where d z It is the difference between adjacent points, indicating the difference between the signal at the zth time point t and the next time point under the detection window, x(t z ) is at the zth time point t z The signal value collected at the moment reflects the operating status of the device at that moment, x(t z+1 ) is at time t z A time point after time t z+1 The collected signal value;

[0117] The difference d between all adjacent points z , calculate its symmetry weight to determine the symmetry characteristics of the signal. The calculation expression is as follows:

[0118]

[0119] Where W s is the symmetry weight, indicating the sth difference value d z The relative proportion among all the difference values, Q is the total number of difference values;

[0120] Symmetry weight W s Represents the difference value d of each adjacent point z The relative proportions within the entire detection window quantify the degree of signal variation at different locations. Specifically, the symmetry weight measures the contribution of each difference value to the total difference, helping to analyze the fluctuation characteristics of the signal at different locations and the symmetry of its distribution. A larger weight indicates more significant fluctuation at that location, potentially indicating asymmetric characteristics or abnormal patterns in the signal. By calculating the symmetry weights of all difference values, we can provide a basis for calculating the skewness index, enabling the fault detection system to more accurately identify the distribution characteristics of the signal and effectively identify potential fault modes.

[0121] Based on the symmetry weight W s , calculate the weighted skewness value of the signal to quantify the symmetry of the data distribution. The calculation expression is as follows:

[0122]

[0123] Where K is the weighted skewness value, which is used to quantify the overall skewness of the signal within the detection window;

[0124] The role of the weighted skewness value K is to quantify the symmetry or skewness of the signal data distribution, thereby helping to identify potential abnormalities or fault features in the signal. By calculating each symmetry weight W s By accumulating the cube of the difference from the uniformly distributed reference value, K can sensitively reflect the asymmetry of the signal distribution within the detection window. If the K value deviates significantly from zero, it indicates that the signal fluctuations are asymmetric, which may indicate the presence of new fault modes that the model has not seen. Conversely, if the K value is close to zero, it indicates a symmetrical signal distribution and a relatively stable operating state, which helps improve the stability and accuracy of fault identification.

[0125] Normalize the weighted skewness value K to generate the skewness index. The calculation expression is as follows:

[0126]

[0127] Where, Skewness Index is the skewness index, max(|K|) is the maximum skewness value observed in the sample data set;

[0128] Within the detection window, the larger the skewness index performance value generated after analyzing the symmetry of the data distribution, the stronger the asymmetry of the data distribution is, which may mean that new fault characteristics have appeared that have not been learned by the model, and therefore are more likely to belong to a fault type that the current model cannot recognize. This is because the trained model usually establishes recognition rules based on the data distribution characteristics of known faults and normal operating conditions, and a large skewness (i.e., a high skewness index value) reflects a significant shift in the data distribution, which is often a manifestation of new fault characteristics. Therefore, when the skewness index reaches a high value, it usually means that the acquired fault characteristics deviate from the recognition range of the model, while a lower skewness index indicates that the data distribution is close to the known pattern learned by the model, indicating that the model can better recognize the current fault characteristics.

[0129] Get the complexity index Complexity Index and Skewness Index After that, the generated complexity index Complexity Index and Skewness Index The data is input into a pre-learned machine learning model, and a fault identification coefficient FIC is generated by the machine learning model. The fault identification coefficient FIC is used to intelligently evaluate the changes in the newly generated fault modes during the operation of the hot stamping equipment.

[0130] The machine learning model is not limited here and can achieve the complexity index Complexity Index and Skewness Index Any machine learning model that generates a fault identification coefficient FIC after comprehensive analysis can be used. In order to implement the technical solution of the present invention, the present invention provides a specific implementation method;

[0131] The formula for generating the fault identification coefficient FIC is as follows:

[0132]

[0133] In the formula, d1 and d2 are the complexity index Complexity Index and Skewness Index The preset proportional coefficient, and d1 and d2 are both greater than 0.

[0134] It can be seen from the calculation expression of the fault identification coefficient that, within the detection window, the larger the complexity index expression value generated after analyzing the complexity of the signal, and the larger the skewness index expression value generated after analyzing the symmetry of the data distribution, the larger the fault identification coefficient expression value generated after the comprehensive analysis of the complexity index and the skewness index, indicating that the current fault feature is a fault feature that the model cannot identify, otherwise it indicates that the fault feature is a recognizable feature that has appeared before.

[0135] A pre-trained machine learning model is a fault detection model that has been established and optimized in advance based on a large amount of historical data and known fault types. Before actual use, the model has been trained using a large amount of sample data (including normal conditions and various fault conditions), and the model's parameters and structure have been adjusted to accurately identify various fault modes in the operation of hot stamping equipment. During the training process, feature data such as the complexity index and skewness index are used as input. The model automatically learns the rules for identifying different fault modes by comparing the feature values ​​in a large amount of sample data with the corresponding equipment status. To improve the model's recognition accuracy and robustness, cross-validation and multiple algorithm testing (such as support vector machines, neural networks, random forests, etc.) are often used for model optimization to ensure that it can maintain high detection accuracy in complex and dynamic equipment operating environments.

[0136] Furthermore, the pre-trained machine learning model has been evaluated and debugged according to established performance standards, enabling it to accurately generate fault identification coefficients during actual operation. When the complexity index and skewness index collected in real time are fed into the model, it calculates the fault identification coefficient for the current device state based on previously learned pattern recognition rules. This coefficient quantifies the degree of abnormality in the device's current operating state. By analyzing changes in the fault identification coefficient, the model can intelligently assess the characteristics of new fault modes and their potential impact on device operation. This evaluation process not only provides timely warnings of potential faults but also adapts to the device's gradually changing operating conditions, thereby improving the recognition rate and detection efficiency of new fault modes for hot stamping equipment, ensuring that the model maintains excellent recognition effectiveness and adaptability during the actual production of hot stamping equipment.

[0137] Based on the results of intelligent evaluation of the machine learning model, the fault types during the operation of the hot stamping equipment are divided into identifiable fault types and unidentifiable fault types;

[0138] The fault identification coefficient generated after comprehensive analysis of the complexity index and skewness index is compared with the pre-set fault identification coefficient reference threshold to classify the fault types during the operation of the hot stamping equipment. The classification steps are as follows;

[0139] If the fault identification coefficient is greater than or equal to a preset fault identification coefficient reference threshold, the fault type during the operation of the hot stamping equipment is classified as an unrecognizable fault type;

[0140] If the fault identification coefficient is less than a preset fault identification coefficient reference threshold, the fault type during the operation of the hot stamping equipment is classified as an identifiable fault type;

[0141] Identifiable fault types are those that the fault detection system can accurately identify and classify. These fault types have typically appeared in the model's training dataset, and the model has learned from this historical data to understand their characteristic patterns and manifestations. Unidentifiable fault types are those that the model has not yet learned, or whose characteristics differ significantly from known fault types. Because these fault patterns do not appear in the training data, the model lacks sufficient knowledge of their characteristics, resulting in an inability to effectively identify or classify them.

[0142] Set a standard model training frequency for identifiable fault types. Regularly train the model to ensure it can adapt to various changes in the equipment, maintain high detection accuracy for known faults, and improve detection accuracy and applicability.

[0143] By setting a standard model training frequency for identifiable fault types, newly collected operating data is regularly added to the training dataset for model updates. This allows the model to learn new data promptly despite changes in equipment operating status and environmental conditions, adapting to normal equipment aging and operating fluctuations, and maintaining high detection accuracy for known fault types. Through continuous model training and optimization, the system not only improves fault detection accuracy but also more effectively adapts to dynamic changes in equipment, ensuring the applicability and stability of the detection model.

[0144] For unrecognizable fault types, the model training frequency is dynamically adjusted based on the standard model training frequency set for recognizable fault types, so that the model can capture the characteristics of new faults in a timely manner and improve the recognition rate;

[0145] For unrecognizable fault types, the model training frequency is dynamically adjusted based on the standard model training frequency set for recognizable fault types, so that the model can capture the characteristics of new faults in a timely manner. The specific steps to improve the recognition rate are as follows:

[0146] The standard model training frequency based on the identifiable fault type is calibrated as F std During operation, the fault identification coefficient FIC is used to quantify the faults detected in the current state of the equipment, that is, by analyzing the characteristics of the equipment operating parameters, it reflects the possible fault intensity or abnormality of the equipment. The fault identification coefficient reference threshold FIC ref It is used to quantify the model's recognition criteria for known faults, that is, as a reference value for judging whether the current fault belongs to a known type. If the fault identification coefficient FIC is greater than or equal to the fault identification coefficient reference threshold FIC ref , it means that there are new, unrecognizable fault types, and the model needs to be dynamically adjusted to improve the ability to identify new faults;

[0147] During detection, the fault identification coefficient FIC is compared with the fault identification coefficient reference threshold FIC ref For comparison:

[0148] If FIC≥FIC ref , then an unrecognizable fault type has occurred, and the model training frequency needs to be dynamically adjusted according to the severity of the fault;

[0149] If FIC <FIC ref , indicating that the fault type can be identified by the current model, there is no need to change the training frequency, and the standard frequency F std ;

[0150] When FIC≥FIC ref , according to the fault identification coefficient FIC for the fault identification coefficient reference threshold FIC refThe training frequency is dynamically adjusted based on the specific percentage of excess, and is calculated using the following formula;

[0151]

[0152] Where, F new is the adjusted training frequency, and H is the adjustment coefficient, which is used to control the increase in training frequency;

[0153] Perform the following operations based on the magnitude by which the fault identification coefficient exceeds the reference threshold:

[0154] If FIC exceeds FIC ref If the percentage is between 0-30%, set H=2 and the training frequency F new Doubled to accommodate the current minor anomaly;

[0155] If FIC exceeds FIC ref If the percentage exceeds 30%, set H=4 and the training frequency F new Improved by three times to accommodate complex fault characteristics and ensure that the model can capture new failure modes in a timely manner;

[0156] With the frequent training of the model, the changes of the fault identification coefficient FIC are monitored in real time. If the fault identification coefficient FIC recovers to a value lower than the fault identification coefficient reference threshold FIC during continuous monitoring, ref , indicating that the model has learned new fault features and restored the training frequency to the standard frequency F std ;

[0157] if FIC <FIC ref ,then F new =F std

[0158] At this point, by restoring the training frequency to the standard frequency, we can maintain the model's efficiency and rational resource utilization, while ensuring the stability of fault detection after the model has adapted to the new features.

[0159] On the basis of dynamically adjusting the model training frequency, continue to monitor the changing trend of the equipment status and fault identification coefficient FIC. If unrecognizable fault types continue to appear frequently, reset the standard frequency F std , to adapt to long-term changes in equipment and improve the adaptability of fault detection models.

[0160] When unrecognizable fault types continue to appear frequently in the equipment operation state, it indicates that the equipment's operating environment or internal state may have undergone continuous and systematic changes. stdThe update rate is no longer sufficient to capture these changes in a timely manner, resulting in an inability to effectively learn new fault signatures. Relying solely on the current standard training frequency, the model may be slow to respond to these frequently occurring anomalies, resulting in missed or false detections and failing to meet the required accuracy for fault detection. Therefore, adjusting the standard training frequency allows the model to acquire new data more frequently, promptly learning the latest operational characteristics of the device, and thus improving its recognition capabilities.

[0161] Adjusting the standard training frequency makes the model update mechanism more flexible and can cope with long-term state changes of the device from the basic frequency. std , the model can only perform short-term emergency updates when specific anomalies occur, and cannot adapt to long-term trend changes in the equipment. By increasing the standard training frequency, the model can continuously self-optimize and adapt to new failure modes and characteristics of the equipment, thereby maintaining the accuracy and stability of fault detection and avoiding the degradation of fault identification performance caused by environmental or state changes.

[0162] The present invention regularly updates the model by setting a standard model training frequency, thereby ensuring that the model can continuously adapt to the normal state changes and known fault modes of the hot stamping equipment. For identifiable fault types, regular model updates can help the model maintain high detection accuracy for these known faults and reduce the occurrence of false positives and missed positives. In a production environment, this strategy can quickly detect known fault modes and issue early warnings in a timely manner to prevent small faults from accumulating into large faults, helping to ensure the stability of the production line and the operating efficiency of the equipment. In addition, this solution establishes a data storage and management system to provide a reliable data foundation for subsequent analysis and feature extraction, allowing the entire fault detection system to be monitored and maintained in real time during equipment operation, reducing production risks caused by equipment downtime or abnormalities.

[0163] The present invention targets unrecognizable fault types, and the solution adopts a strategy of dynamically adjusting the frequency of model training so that the model can capture and learn the characteristics of new faults in a timely manner. When the fault recognition coefficient reaches or exceeds the reference threshold, the system will identify the fault as an unrecognizable fault, thereby adding new data to the training set through higher-frequency training, prompting the model to quickly adapt to new fault modes. This not only enables the timely identification of new fault modes, but also effectively improves the adaptability and recognition rate of the model. Through this strategy of dynamically adjusting the training frequency, the risk of production downtime and equipment damage caused by new faults not being detected in a timely manner can be significantly reduced, and problems can be prevented from accumulating during equipment operation to the point of endangering the stability of the production line or even causing production stoppages, thereby improving production efficiency and product quality reliability.

[0164] The present invention provides Figure 2The hot stamping equipment fault detection system shown here, based on data collection and analysis, includes a data collection and management module, a feature extraction and intelligent evaluation module, a fault type classification module, a recognizable fault periodic update module, and an unrecognizable fault dynamic adaptation module:

[0165] The data collection and management module first obtains the various operating parameter information generated during the operation of the hot stamping equipment. At the same time, it establishes a data storage and management system to effectively save and organize the large amount of collected data for subsequent retrieval and processing.

[0166] The feature extraction and intelligent evaluation module extracts features from pre-processed hot stamping equipment operating parameter data. Based on the extracted fault feature information, it uses a pre-learned machine learning model to intelligently evaluate new fault mode changes generated during the hot stamping equipment operation.

[0167] The fault type classification module, based on the results of intelligent evaluation of the machine learning model, divides the fault types during the operation of the hot stamping equipment into identifiable fault types and unidentifiable fault types;

[0168] The identifiable fault regular update module sets a standard model training frequency for identifiable fault types. By regularly training the model, it ensures that it can adapt to various changes in the equipment, maintains high detection accuracy for known faults, and improves detection accuracy and applicability.

[0169] The unrecognizable fault dynamic adaptation module dynamically adjusts the model training frequency for unrecognizable fault types based on the standard model training frequency set for recognizable fault types, so that the model can capture the characteristics of new faults in a timely manner and improve the recognition rate.

[0170] An embodiment of the present invention provides a hot stamping equipment fault detection method based on data collection and analysis, which is implemented by the above-mentioned hot stamping equipment fault detection system based on data collection and analysis. The specific method and process of the hot stamping equipment fault detection system based on data collection and analysis are detailed in the embodiment of the above-mentioned hot stamping equipment fault detection method based on data collection and analysis, and will not be repeated here.

[0171] The above description is merely illustrative of certain exemplary embodiments of the present invention. It goes without saying that those skilled in the art will be able to modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and description are illustrative in nature and should not be construed as limiting the scope of protection of the claims.

Claims

1. A method for detecting faults in hot stamping equipment based on data collection and analysis, characterized in that: The following steps are involved: First, obtain the various operating parameter information generated during the operation of the hot stamping equipment. At the same time, establish a data storage and management system to effectively save and organize the large amount of collected data to facilitate subsequent call and processing; Perform feature extraction on pre-processed hot stamping equipment operating parameter data. Based on the extracted fault feature information, a pre-learned machine learning model is used to intelligently evaluate the changes in new fault modes generated during the operation of the hot stamping equipment. Based on the results of intelligent evaluation of the machine learning model, the fault types during the operation of the hot stamping equipment are divided into identifiable fault types and unidentifiable fault types; Set a standard model training frequency for identifiable fault types. Regularly train the model to ensure it can adapt to various changes in the equipment, maintain high detection accuracy for known faults, and improve detection accuracy and applicability. For unrecognizable fault types, the model training frequency is dynamically adjusted based on the standard model training frequency set for recognizable fault types, so that the model can capture the characteristics of new faults in a timely manner and improve the recognition rate.

2. The method for detecting hot stamping equipment faults based on data collection and analysis according to claim 1, characterized in that: The steps to obtain the various operating parameter information generated during the operation of the hot stamping equipment and establish a data storage and management system to effectively save and organize the large amount of collected data for subsequent call and processing are as follows: After obtaining various parameters during the operation of the hot stamping equipment, data collection and real-time transmission are first performed; Clean and format the collected data to ensure data quality and consistency; Store cleaned and formatted data in a database and establish an efficient storage and indexing system to ensure that data can be quickly retrieved and managed; Establish a complete data management system and set access permissions to ensure that data is effectively protected while being used efficiently.

3. The method for detecting hot stamping equipment faults based on data collection and analysis according to claim 1, characterized in that: Feature extraction is performed on the pre-processed hot stamping equipment operating parameter data. The extracted features include the complexity of the signal and the symmetry of the data distribution. After acquisition, the complexity of the signal and the symmetry of the data distribution are analyzed under the detection window to generate the complexity index and skewness index respectively. The complexity index quantifies the complexity of the signal, that is, the diversity and instability of the signal changes in the time or frequency domain, reflecting the fluctuation amplitude and uncertainty of the equipment operation status. The skewness index quantifies the symmetry of the data distribution, that is, whether the data is symmetrically distributed around the average value, reflecting the degree of offset and abnormal deviation trend in the equipment status data.

4. The method for detecting hot stamping equipment faults based on data collection and analysis according to claim 3, characterized in that: After obtaining the complexity index and skewness index, the generated complexity index and skewness index are input into the pre-learned machine learning model, and the fault identification coefficient is generated by the machine learning model. The fault identification coefficient is used to perform an intelligent evaluation of the newly generated fault mode changes during the operation of the hot stamping equipment.

5. The method for detecting hot stamping equipment failure based on data collection and analysis according to claim 4, characterized in that: The fault identification coefficient generated after comprehensive analysis of the complexity index and skewness index is compared with the pre-set fault identification coefficient reference threshold to classify the fault types during the operation of the hot stamping equipment. The classification steps are as follows; If the fault identification coefficient is greater than or equal to a preset fault identification coefficient reference threshold, the fault type during the operation of the hot stamping equipment is classified as an unrecognizable fault type; If the fault identification coefficient is less than a preset fault identification coefficient reference threshold, the fault type during the operation of the hot stamping equipment is classified as an identifiable fault type.

6. The method for detecting hot stamping equipment faults based on data collection and analysis according to claim 5, characterized in that: For unrecognizable fault types, the model training frequency is dynamically adjusted based on the standard model training frequency set for recognizable fault types, so that the model can capture the characteristics of new faults in a timely manner. The specific steps to improve the recognition rate are as follows: The standard model training frequency based on the identifiable fault type is calibrated as F std During operation, the fault identification coefficient FIC is used to quantify the faults detected in the current state of the equipment. The fault identification coefficient reference threshold FIC ref It is used to quantify the model's recognition criteria for known faults. If the fault recognition coefficient FIC is greater than or equal to the fault recognition coefficient reference threshold FIC ref , it means that there are new, unrecognizable fault types, and the model needs to be dynamically adjusted to improve the ability to identify new faults; During detection, the fault identification coefficient FIC is compared with the fault identification coefficient reference threshold FIC ref For comparison: If FIC≥FIC ref , then an unrecognizable fault type has occurred, and the model training frequency needs to be dynamically adjusted according to the severity of the fault; If FIC <FIC ref , indicating that the fault type can be identified by the current model, there is no need to change the training frequency, and the standard frequency F std ; When FIC≥FIC ref , according to the fault identification coefficient FIC for the fault identification coefficient reference threshold FIC ref The training frequency is dynamically adjusted based on the specific percentage of excess, and is calculated using the following formula; Where, F new is the adjusted training frequency, and H is the adjustment coefficient, which is used to control the increase in training frequency; Perform the following operations based on the magnitude by which the fault identification coefficient exceeds the reference threshold: If FIC exceeds FIC ref If the percentage is between 0-30%, set H=2 and the training frequency F new Doubled to accommodate the current minor anomaly; If FIC exceeds FIC ref If the percentage exceeds 30%, set H=4 and the training frequency F new Improved by three times to accommodate complex fault characteristics and ensure that the model can capture new failure modes in a timely manner; With the frequent training of the model, the changes of the fault identification coefficient FIC are monitored in real time. If the fault identification coefficient FIC recovers to a value lower than the fault identification coefficient reference threshold FIC during continuous monitoring, ref , indicating that the model has learned new fault features and restored the training frequency to the standard frequency F std ; if FIC<FIC ref ,then F new =F std At this point, by restoring the training frequency to the standard frequency, we can maintain the model's efficiency and rational resource utilization, while ensuring the stability of fault detection after the model has adapted to the new features. On the basis of dynamically adjusting the model training frequency, continue to monitor the changing trend of the equipment status and fault identification coefficient FIC. If unrecognizable fault types continue to appear frequently, reset the standard frequency F std , to adapt to long-term changes in equipment and improve the adaptability of fault detection models.

7. The method for detecting faults of hot stamping equipment based on data collection and analysis according to claim 3, characterized in that: In the detection window, the complexity of the signal is analyzed and the specific steps to generate the complexity index are as follows: In the detection window, the pre-processed hot stamping equipment operation signal is subjected to wavelet transform to decompose it into time-frequency features of different scales and frequency components. The calculation expression is as follows: Where W x (a, b) are the wavelet transform coefficients, i.e., the signal characteristics captured by the pre-processed hot stamping equipment operation signal x(t) at a given scale a and position b. Scale a is the scale factor, which is used to control the expansion and contraction of the mother wavelet. Position b is the translation factor, which is used to control the position of the mother wavelet on the time axis, indicating the local change of the signal during this period. * is the complex conjugate of the mother wavelet, ψ is the wavelet function used to decompose the signal; For the wavelet transform coefficient W x (a, b) Calculate the sum of squares to calculate the energy distribution of the signal at each scale. The calculation expression is as follows: Where E(a) is the energy distribution, that is, the energy distribution at scale a; The sample entropy is calculated at each scale to quantify the randomness and complexity of the signal at that scale. The calculation expression is as follows: Where, is the sample entropy, m is the embedding dimension, r is the similarity tolerance, N is the total number of samples, which defines the length of the entire signal and the analysis range, and i represents the index position of the sample, that is, the position of each data point in the signal. is the number of similar patterns of the i-th sample under the embedding dimension m and similarity tolerance r, is the number of similar patterns of the i-th sample under the embedding dimension m+1 and similarity tolerance r; The normalized energy distribution of each scale is combined with the corresponding sample entropy to generate a complexity index to quantify the overall complexity of the signal. The calculation expression is as follows: Where Complexity Index is the complexity index, is the sample entropy on scale a.

8. The method for detecting faults of hot stamping equipment based on data collection and analysis according to claim 3, characterized in that: Under the detection window, the symmetry of the data distribution is analyzed and the specific steps to generate the skewness index are as follows: In the detection window, the pre-processed hot stamping equipment operation signal is calibrated as x(t), and the difference between each sampling point and its adjacent points is calculated. The expression is as follows: d z =|x(t z )-x(t z+1 )| Where, d z is the difference between adjacent points, x(t z ) is at the zth time point t z The signal value collected at the moment, x(t z+1 ) is at time t z A time point after time t z+1 The collected signal value; The difference d between all adjacent points z , calculate its symmetry weight to determine the symmetry characteristics of the signal. The calculation expression is as follows: Where W s is the symmetry weight, Q is the total number of differences; Based on the symmetry weight W s , calculate the weighted skewness value of the signal to quantify the symmetry of the data distribution. The calculation expression is as follows: Where K is the weighted skewness value; Normalize the weighted skewness value K to generate the skewness index. The calculation expression is as follows: Where, Skewness Index is the skewness index, and max(|K|) is the maximum skewness value observed in the sample data set.

9. A hot stamping equipment fault detection system based on data collection and analysis, used to implement the hot stamping equipment fault detection method based on data collection and analysis as described in any one of claims 1 to 8, characterized in that: It includes data collection and management module, feature extraction and intelligent evaluation module, fault type classification module, identifiable fault periodic update module and unidentifiable fault dynamic adaptation module: The data collection and management module first obtains the various operating parameter information generated during the operation of the hot stamping equipment. At the same time, it establishes a data storage and management system to effectively save and organize the large amount of collected data for subsequent retrieval and processing. The feature extraction and intelligent evaluation module extracts features from pre-processed hot stamping equipment operating parameter data. Based on the extracted fault feature information, it uses a pre-learned machine learning model to intelligently evaluate new fault mode changes generated during the hot stamping equipment operation. The fault type classification module, based on the results of intelligent evaluation of the machine learning model, divides the fault types during the operation of the hot stamping equipment into identifiable fault types and unidentifiable fault types; The identifiable fault regular update module sets a standard model training frequency for identifiable fault types. By regularly training the model, it ensures that it can adapt to various changes in the equipment, maintains high detection accuracy for known faults, and improves detection accuracy and applicability. The unrecognizable fault dynamic adaptation module dynamically adjusts the model training frequency for unrecognizable fault types based on the standard model training frequency set for recognizable fault types, so that the model can capture the characteristics of new faults in a timely manner and improve the recognition rate.

Images