A fault diagnosis method and system for a punch safety guard

Abstract

This invention relates to the field of industrial automation operation and maintenance technology, and discloses a fault diagnosis method and system for a stamping safety and anti-pressure device. The method includes collecting multi-source data on device operating status, load, and component wear; obtaining a purified signal through multi-scale wavelet decomposition and threshold shrinkage denoising; extracting time-frequency joint features to construct a high-dimensional vector; determining anomalies through dual thresholds and matching historical fault patterns; locating abnormal energy accumulation intervals and predicting deviation trends and fluctuation amplitudes; and fusing load and wear data to perform polynomial regression to fit a high-confidence fault evolution trajectory. This method enables early and accurate fault diagnosis of stamping safety and anti-pressure devices, meeting the core requirements of industrial automation operation and maintenance for production safety and process stability.

Description

Technical Field

[0001] This invention relates to the field of industrial automation operation and maintenance technology, and in particular to a fault diagnosis method and system for a stamping safety anti-pressure device. Background Technology

[0002] Currently, in the field of industrial automation operation and maintenance, with the continuous expansion of manufacturing production scale and the increasing demand for precision stamping processes, stamping safety and anti-pressure devices, as core equipment to ensure production safety, are directly related to personnel operation safety and production process continuity through their stable operation.

[0003] Current fault diagnosis of stamping safety and anti-pressure devices in the industry mainly relies on regular manual inspections or monitoring of single physical parameters. For example, this involves acquiring equipment operating data using handheld vibration detectors, judging signal anomalies using fixed thresholds, or ignoring the correlation analysis of multi-source data. However, this approach is clearly insufficient in complex operating environments. Manual inspections have limited coverage, inherent detection delays, and are susceptible to errors due to differences in the experience of maintenance personnel, easily missing hidden faults such as minor off-center loads. Single-parameter monitoring cannot accurately distinguish between normal equipment fluctuations and fault characteristics, and lacks the multi-source data collaborative analysis capabilities provided by IoT chips, resulting in a high misjudgment rate, especially in high-load, high-frequency stamping scenarios, making it difficult to quickly locate the source of the fault.

[0004] In summary, existing technologies are insufficient for accurate early fault diagnosis of stamping safety and anti-pressure devices, and cannot meet the core requirements of industrial automation operation and maintenance for production safety and process stability. Summary of the Invention

[0005] This invention provides a fault diagnosis method and system for a stamping safety anti-pressure device, so as to achieve accurate early fault diagnosis of the stamping safety anti-pressure device and meet the core requirements of industrial automation operation and maintenance for production safety and process stability.

[0006] In a first aspect, to solve the above-mentioned technical problems, the present invention provides a fault diagnosis method for a stamping safety anti-pressure device, comprising: Acquire operating status data, load data, and component wear data of the stamping safety anti-pressure device; The operating status data is denoised to obtain a denoised signal sequence; Extract the time-domain statistical features and frequency-domain energy features of the denoised signal sequence, construct a feature vector, and analyze the fluctuation amplitude and time-domain variation law of the feature vector under different time windows; If the fluctuation amplitude exceeds a preset amplitude judgment threshold, and the time domain change pattern exceeds a preset time series judgment threshold, then the feature vector is matched with a pre-acquired historical anomaly pattern to obtain an anomaly matching result. Based on the anomaly matching results, the distribution characteristics of the frequency domain energy of the corresponding signal segment are analyzed, and a time-series feature sequence is formed to locate the abnormal energy accumulation region in the features and determine the abnormal distribution interval. Calculate the deviation between the signal and the preset historical baseline data within the distribution interval, and analyze the trend of the deviation within the prediction time window; if the trend of the deviation conforms to the preset abnormal evolution pattern, it is determined that the abnormal evolution condition is met. Based on the abnormal evolution conditions, the load data, and the component wear data, a preliminary fault evolution path is fitted using polynomial regression, and the prediction confidence is evaluated. If the prediction confidence exceeds a preset confidence threshold, the fitted path is determined as the fault evolution trajectory. Based on the fault evolution trajectory, a data acquisition strategy, fault early warning threshold, and monitoring cycle are formulated to generate a precise monitoring scheme for the stamping safety anti-pressure device.

[0007] Secondly, the present invention provides a fault diagnosis system for a stamping safety anti-pressure device, comprising: The data acquisition module is used to acquire the operating status data, load data, and component wear data of the stamping safety anti-pressure device; A signal denoising module is used to denoise the operating status data to obtain a denoised signal sequence. The feature extraction module is used to extract the time-domain statistical features and frequency-domain energy features of the denoised signal sequence, construct a feature vector, and analyze the fluctuation amplitude and time-domain variation law of the feature vector under different time windows. An anomaly identification module is used to match the feature vector with a pre-acquired historical anomaly pattern to obtain an anomaly matching result if the fluctuation amplitude exceeds a preset amplitude judgment threshold and the time domain change pattern exceeds a preset time series judgment threshold. The interval positioning module is used to analyze the frequency domain energy distribution characteristics of the corresponding signal segment based on the anomaly matching results, form a time-series feature sequence, locate the abnormal energy accumulation area in the features, and determine the abnormal distribution interval. The evolution determination module is used to calculate the deviation value between the signal in the distribution interval and the preset historical baseline data, and analyze the changing trend of the deviation value within the prediction time window; if the changing trend conforms to the preset abnormal evolution mode, it is determined that the abnormal evolution condition is met. The trajectory fitting module is used to fit a preliminary fault evolution path through polynomial regression based on the abnormal evolution conditions, the load data and the component wear data, and to evaluate the prediction confidence. If the prediction confidence exceeds a preset confidence threshold, the fitted path is determined as the fault evolution trajectory. The scheme generation module is used to formulate data acquisition strategies, fault early warning thresholds and monitoring cycles based on the fault evolution trajectory, and generate a precise monitoring scheme for the stamping safety anti-pressure device.

[0008] Compared with the prior art, the present invention has the following beneficial effects: (1) This invention obtains the operating status data, load data and component wear data of the stamping safety anti-pressure device, performs multi-scale decomposition and threshold shrinkage noise reduction on the sensor signal, extracts time domain statistical features and frequency domain energy features to construct vectors, breaks through the limitations of traditional manual inspection with limited coverage and reliance on experience, explores the multi-dimensional correlation features of equipment operation, eliminates interference from environmental noise and operational differences, provides high-precision basic data support for fault diagnosis, effectively improves the capture rate of hidden faults such as minor off-center loads, and solves the problem of manual missed detection of early weak anomalies.

[0009] (2) This invention analyzes the fluctuation amplitude and time domain change law of feature vector in different time windows, compares the historical abnormal patterns to match potential faults after double threshold judgment, locates the abnormal energy accumulation area to determine the signal distribution range, breaks through the limitation of traditional single parameter that cannot distinguish between normal fluctuations and faults, accurately captures the exclusive evolution characteristics of stamping equipment faults, provides multi-dimensional basis for anomaly judgment, significantly improves the accuracy of fault identification under complex working conditions, and makes up for the high misjudgment rate of existing technology.

[0010] (3) This invention calculates the trend of signal deviation value change, fits the fault evolution trajectory by combining load data and component wear data, formulates dynamic sampling strategy, multi-level early warning threshold and monitoring cycle, breaks through the limitations of traditional lack of dynamic evolution analysis and accurate monitoring scheme, provides full-cycle fault control basis for operation and maintenance, solves the problems of difficult fault location and lagging monitoring, takes into account the accuracy of diagnosis and production safety, and meets the core requirements of industrial automation operation and maintenance for process stability. Attached Figure Description

[0011] Figure 1 This is a schematic flowchart of a fault diagnosis method for a stamping safety anti-pressure device provided in the first embodiment of the present invention; Figure 2 This is a schematic diagram of the fault diagnosis system of a stamping safety anti-pressure device provided in the second embodiment of the present invention. Detailed Implementation

[0012] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0013] Reference Figure 1 The first embodiment of the present invention provides a fault diagnosis method for a stamping safety anti-pressure device, comprising the following steps: S101, acquire the operating status data, load data and component wear data of the stamping safety anti-pressure device; S102, Denoise the running status data to obtain a denoised signal sequence; S103, extract the time-domain statistical features and frequency-domain energy features of the denoised signal sequence, construct a high-dimensional feature vector, and analyze the fluctuation amplitude and trend slope of the high-dimensional feature vector under different time windows; S104, if the fluctuation amplitude exceeds a preset amplitude judgment threshold and the trend slope exceeds a preset slope judgment threshold, then the high-dimensional feature vector is matched with the pre-acquired historical anomaly pattern to obtain an anomaly matching result. S105, perform frequency domain analysis based on the anomaly matching results to obtain a time-series feature sequence, locate the anomaly clustering region based on the time-series feature sequence, and determine the anomaly distribution interval; S106, extract the signal feature sequence within the abnormal distribution interval, and make predictions based on the signal feature sequence to obtain the predicted trend slope and predicted fluctuation amplitude; S107, perform polynomial regression fitting based on the predicted trend slope, predicted fluctuation amplitude, load data, and component wear data to obtain the fault evolution trajectory.

[0014] In step S101, acquiring the operating status data, load data, and component wear data of the stamping safety anti-pressure device includes: A sensor array is deployed to collect raw data on the operating status of the stamping safety and anti-pressure device. The raw data on the operating status includes vibration signals, pressure signals, and photoelectric sensing signals. The real-time load data of the stamping safety anti-pressure device is read through the equipment controller; The wear monitoring unit collects cumulative wear data of components. The original operating status data, the real-time load data, and the cumulative wear data of the components are respectively processed by format standardization to obtain operating status data, load data, and component wear data.

[0015] It should be noted that, firstly, a sensor array is deployed to collect raw data on the operating status of the stamping safety and anti-pressure device. Data acquisition relies on an array composed of piezoelectric vibration sensors, diffused silicon pressure sensors, and through-beam photoelectric sensing units. Vibration signals, pressure signals, and photoelectric sensing signals are simultaneously collected throughout the entire operation of the equipment. The sampling frequency is set according to the dynamic characteristics of the stamping process. The sampling frequency threshold is set based on statistical data of high-frequency operation signals from stamping equipment over the past year. The basic sampling frequency is set to 2048 Hz, which can be increased to 4096 Hz for high-precision stamping scenarios and decreased to 1024 Hz for conventional stamping scenarios. Those skilled in the art can adjust this threshold within the range of 1024 to 4096 Hz according to the stamping cycle speed.

[0016] For example, in an automotive parts stamping production line, multiple types of sensor arrays are used to synchronously collect slider vibration, die pressure, and safety light curtain photoelectric signals to form a complete set of raw data on the operating status.

[0017] The equipment controller reads the real-time load data of the stamping safety anti-pressure device. The data reading interface is connected to the programmable logic controller of the stamping equipment to obtain three types of load parameters in real time: stamping stroke load, motor output power and slide impact load. The reading cycle is synchronized with the sensor sampling cycle to ensure data timing alignment. All raw load data is directly retained for subsequent standardization processing.

[0018] For example, the device controller can synchronously read the real-time impact load data of the stamping equipment five times per second to obtain the original load data that matches the timing of the sensor signal.

[0019] Next, the wear monitoring unit collects the cumulative wear data of the components. The wear monitoring adopts a contact displacement sensing monitoring unit to continuously collect the cumulative wear displacement of the mold cutting edge, slide guide rail and safety protection link. The cumulative statistical period is based on a single equipment start-up operation, and the cumulative change value of component wear is recorded throughout the entire cycle, with no data loss.

[0020] For example, a displacement monitoring unit can be used to continuously collect wear displacement data of the die cutting edge and record the total wear of parts in a single production batch.

[0021] Subsequently, the raw operational status data, real-time load data, and cumulative component wear data were standardized to obtain operational status data, load data, and component wear data, respectively. The standardization process unified the timestamp format, physical dimensions, and data storage structure of all data. All numerical parameters were mapped to the range of 0 to 1 using the minimum-maximum normalization method, eliminating the magnitude differences and format barriers between different data types. The standardized data can then be directly used for subsequent signal processing and feature analysis.

[0022] For example, vibration amplitude, pressure values ​​and wear displacement are dimensionally unified and normalized to generate standardized operating status data, load data and component wear data with consistent format.

[0023] In step S102, the denoising process performed on the operating status data to obtain a denoised signal sequence includes: Extract the fluctuation characteristics of the operating status data, and determine the level depth of the multi-scale decomposition based on the fluctuation characteristics; The running status data is decomposed based on the level depth to obtain high-frequency detail components and low-frequency approximation components; The high-frequency detail components are subjected to spectral analysis by fast Fourier transform to obtain the high-frequency band corresponding to the noise, and the total signal energy within the high-frequency band is calculated to obtain the noise energy value. If the noise energy value exceeds the preset noise judgment threshold, then the high-frequency detail component is subjected to threshold shrinkage processing to obtain the optimized high-frequency component; The optimized high-frequency components and the low-frequency approximation components are reconstructed to obtain a denoised signal sequence.

[0024] It should be noted that, firstly, the fluctuation characteristics of the operating status data are extracted, and the depth of the multi-scale decomposition is determined based on these characteristics. The fluctuation characteristics are obtained by calculating the signal variance and waveform kurtosis. The multi-scale decomposition employs discrete wavelet transform technology, with the db4 wavelet as the wavelet basis function. The decomposition depth is set based on the frequency band distribution data of the stamping equipment's operating signals over the past year. The basic level is set to 4 levels, which can be increased to 5 levels for high-precision fault monitoring scenarios and decreased to 3 levels for conventional stamping operation scenarios. This level division adapts to the multi-frequency characteristics of stamping signals and can stably separate effective signals from interference components.

[0025] For example, the kurtosis value of the vibration signal of the stamping slider is calculated to be 3.2, and the decomposition level of the discrete wavelet transform is determined to be 4 levels based on the fluctuation amplitude.

[0026] By decomposing the operational status data according to the hierarchical depth, high-frequency detail components and low-frequency approximate components are obtained. Discrete wavelet transform completes the layer-by-layer decomposition according to the set hierarchy. The low-frequency approximate components carry the basic trend signal of equipment operation, while the high-frequency detail components contain transient impacts and environmental interference noise. All components maintain the original temporal arrangement, and the data dimensions are completely matched with the original signal. For example, performing a 4-level wavelet decomposition on the standardized vibration signal yields a set of low-frequency approximate components and four sets of independent high-frequency detail components.

[0027] Next, a spectral analysis of the high-frequency detail components is performed using Fast Fourier Transform (FFT) to obtain the high-frequency band corresponding to the noise. The total signal energy within the high-frequency band is then calculated to obtain the noise energy value. FFT is used to perform frequency domain transformation on each layer of high-frequency components, locking the noise frequency band in the high-frequency range above 1000 Hz. Energy statistics cover all sampling points across the entire frequency band, and all energy values ​​are mapped to the 0-1 range using the min-max normalization method. For example, after performing frequency domain transformation on the high-frequency components, the noise frequency band above 1200 Hz is locked, and the normalized noise energy value is calculated to be 0.72.

[0028] It is worth noting that the noise judgment threshold is set based on statistical data of on-site interference in the stamping workshop over the past year. The basic threshold is set to 0.5, which can be increased to 0.6 under strong electromagnetic interference conditions and decreased to 0.4 in low-interference clean production scenarios. Those skilled in the art can adjust this threshold within the range of 0.4 to 0.6 according to the intensity of interference in the on-site environment. If the noise energy value exceeds the preset noise judgment threshold, soft threshold shrinkage processing is performed on the high-frequency detail components to obtain optimized high-frequency components. The processing can smooth the signal waveform and avoid reconstruction distortion problems.

[0029] Finally, the optimized high-frequency components and low-frequency approximate components are reconstructed to obtain a denoised signal sequence. Signal reconstruction employs inverse discrete wavelet transform (IDE) technology, fusing all component data in reverse order according to the decomposition level. The reconstructed signal fully retains the effective characteristics of equipment operation, eliminates mechanical vibration and electromagnetic interference noise, and maintains precise alignment between the sequence timing and the original operating data. For example, fusing the optimized high-frequency components and the original low-frequency approximate components to perform an inverse wavelet transform generates an interference-free denoised signal sequence for the stamping equipment operation.

[0030] In step S103, the extraction of time-domain statistical features and frequency-domain energy features of the denoised signal sequence, and the construction of a high-dimensional feature vector, along with the analysis of the fluctuation amplitude and trend slope of the high-dimensional feature vector under different time windows, includes: A fixed-length sliding window is set to slice the denoised signal sequence to obtain multiple continuous signal segments; The mean, variance, peak factor, and kurtosis of each signal segment are calculated to obtain the time-domain statistical characteristics; Perform spectral analysis on each signal segment to extract the energy proportion, spectral peak value, and harmonic components of the characteristic frequency band, and obtain the frequency domain energy characteristics. The time-domain statistical features and the frequency-domain energy features are concatenated according to their dimensions to construct a high-dimensional feature vector; The Euclidean distance of the high-dimensional feature vectors corresponding to the continuous time window is calculated to obtain the fluctuation amplitude; By fitting the values ​​of each dimension of the high-dimensional feature vector across multiple consecutive time windows, the slope of the trend of each feature dimension changing over time can be obtained.

[0031] It should be noted that, firstly, a fixed-length sliding window is set to slice the denoised signal sequence, resulting in multiple continuous signal segments. The sliding window parameters are set based on the timing characteristics statistics of stamping equipment operation signals over the past year. The basic window length is set to 50 milliseconds, and the sliding step size is set to 10 milliseconds. For high-precision fault capture scenarios, the window length can be shortened to 30 milliseconds, while for low-speed stamping scenarios, the window length can be extended to 80 milliseconds. The slicing process maintains the continuity of the signal timing, with no data overlap or loss. For example, using a 50-millisecond window and a 10-millisecond step size to slice the light curtain denoised signal generates multiple sets of continuous and non-overlapping short signal segments.

[0032] Next, the mean, variance, peak factor, and kurtosis of each signal segment are calculated to obtain the time-domain statistical characteristics. The four types of time-domain indicators respectively characterize the baseline level, dispersion, impulse intensity, and distribution pattern of the signal. All calculated values ​​are mapped to the interval between 0 and 1 using the min-max normalization method to eliminate the difference in numerical magnitude between different segments. The feature data can be directly used for subsequent vector construction.

[0033] For example, four time-domain indices are calculated for a single signal segment, and after normalization, a standardized set of time-domain statistical features is obtained.

[0034] Subsequently, spectral analysis was performed on each signal segment to extract the energy proportion, spectral peak value, and harmonic components of the characteristic frequency band, thus obtaining the frequency domain energy characteristics. The spectral analysis employed short-time Fourier transform technology, locking the characteristic frequency band within the 10 to 500 Hz range sensitive to stamping faults. All frequency domain values ​​underwent synchronous normalization processing to maintain a consistent data dimensionality standard with the time domain characteristics. For example, a frequency domain transform was performed on the signal segment to extract the energy proportion and spectral peak value of the target frequency band, generating standardized frequency domain energy characteristic data.

[0035] Time-domain statistical features and frequency-domain energy features are concatenated dimensionally to construct a high-dimensional feature vector. The vector dimension is fixed at seven dimensions, arranged in the following order: mean, variance, peak factor, kurtosis, energy percentage, spectral peak value, and harmonic components. All features are concatenated sequentially to form a one-dimensional vector, and each signal segment corresponds to a unique high-dimensional feature vector. For example, concatenating four time-domain features and three frequency-domain features of a single segment generates a standardized seven-dimensional high-dimensional feature vector.

[0036] The fluctuation amplitude is obtained by calculating the Euclidean distance between the high-dimensional feature vectors corresponding to continuous time windows. The Euclidean distance quantifies the degree of feature difference between adjacent window vectors; a higher value indicates a greater fluctuation amplitude in the signal state. The calculation results are normalized and mapped to the interval between 0 and 1 to standardize the evaluation criteria for fluctuation degree. For example, calculating the spatial distance between two adjacent sets of feature vectors, after normalization, yields a signal fluctuation amplitude of 0.18.

[0037] Finally, the values ​​of each dimension of the high-dimensional feature vector across multiple consecutive time windows are fitted to obtain the trend slope of each feature dimension over time. The fitting operation employs a linear regression algorithm, performing time-series fitting on each of the seven features one by one. The slope value represents the rate of change of the corresponding feature; a positive value indicates an increasing feature, and a negative value indicates a decreasing feature. The fitting result is free of distortion or bias. For example, performing linear fitting on the variance features of ten consecutive windows yields a positive trend slope of 0.22, indicating a continuous increase in the signal dispersion.

[0038] In step S104, if the fluctuation amplitude exceeds a preset amplitude determination threshold and the trend slope exceeds a preset slope determination threshold, then the high-dimensional feature vector is matched with a pre-acquired historical anomaly pattern to obtain an anomaly matching result, including: The fluctuation amplitude is compared with a preset amplitude judgment threshold, and the trend slope is compared with a preset slope judgment threshold. If the comparison results all exceed the corresponding threshold, then the signal segment within the corresponding time window is extracted as the abnormal sequence to be detected. Calculate the difference between the feature distribution of the abnormal sequence to be detected and the feature distribution of the historical abnormal patterns obtained in advance; If the difference is lower than the preset pattern matching threshold, the detected abnormal sequence is determined to match the historical abnormal pattern, and an abnormal matching result is obtained.

[0039] It should be noted that, firstly, the fluctuation amplitude is compared with a preset amplitude judgment threshold, and simultaneously, the trend slope is compared with a preset slope judgment threshold. The amplitude judgment threshold is set based on the signal fluctuation statistics of the stamping equipment during normal operation over the past year. The basic threshold is set to 0.3, which can be lowered to 0.2 for high-precision fault protection scenarios and raised to 0.4 for conventional stamping production scenarios. Those skilled in the art can adjust this threshold within the range of 0.2 to 0.4 according to the equipment protection level. The slope judgment threshold is set synchronously based on the time-series change pattern, with a basic threshold set to 0.15 to adapt to the abnormal trend judgment needs of all scenarios. For example, if the current signal fluctuation amplitude is 0.42 and the trend slope is 0.21, both values ​​exceed the corresponding set judgment thresholds.

[0040] If the comparison results all exceed the corresponding thresholds, the signal segment within the corresponding time window is extracted as the sequence of mutations to be detected. The extraction process strictly aligns the signal time nodes, fully preserving the time and frequency domain features within that time period, without data truncation or feature loss, thus providing a complete feature data foundation for subsequent pattern matching. For example, a 50-millisecond signal window with both thresholds exceeding the limit is locked, and the complete signal data for the corresponding time period is extracted to generate the sequence of mutations to be detected.

[0041] It is worth noting that the difference between the feature distribution of the detected abnormal sequence and the feature distribution of pre-acquired historical abnormal patterns is calculated using a dynamic time warping algorithm. This algorithm quantifies the morphological differences between the two types of sequences in the feature space, and all difference values ​​are mapped to the interval between 0 and 1 using a minimum-maximum normalization method. The pattern matching threshold is set based on the statistical matching accuracy of historical fault samples. The basic threshold is set to 0.4, which can be lowered to 0.3 for high matching accuracy scenarios and raised to 0.5 for lenient judgment scenarios. For example, the difference between the detected sequence and the abnormal pattern of microcracks in the mold, after normalization, yields a difference value of 0.32.

[0042] If the difference is lower than the preset pattern matching threshold, the abnormal sequence to be detected is determined to match the historical abnormal pattern, and the abnormal matching result is obtained. The matching result is synchronously associated with the fault type label, which clarifies the equipment fault mode corresponding to the current abnormality, and provides a direct basis for subsequent fault location and evolution analysis.

[0043] For example, if the difference is 0.32, which is lower than the pattern matching threshold of 0.4, the current sequence is determined to match the abnormal pattern of microcracks in the mold, and the corresponding abnormal matching result is generated.

[0044] In step S105, the step of performing frequency domain analysis based on the anomaly matching results to obtain a time-series feature sequence, locating anomaly clustering regions based on the time-series feature sequence, and determining anomaly distribution intervals includes: Based on the anomaly matching results, extract the frequency domain energy distribution data of the corresponding signal segment; The frequency domain energy distribution data are integrated in chronological order to form a temporal feature sequence; The marker points in the time-series feature sequence that exceed a preset energy threshold are selected, and the time-domain interval and frequency band range in which the marker points appear consecutively are identified as abnormal energy accumulation areas. Calculate the start and end positions of the abnormal energy accumulation region in the original signal to determine the abnormal distribution range.

[0045] It should be noted that, firstly, based on the anomaly matching results, the frequency domain energy distribution data of the corresponding signal segments are extracted. Frequency domain analysis employs short-time Fourier transform technology, focusing on the 10 to 500 Hz frequency band sensitive to stamping faults to complete data extraction. All energy values ​​are mapped to the 0-1 range using the minimum-maximum normalization method, ensuring the data range fully covers the full frequency band information corresponding to the anomaly signal. For example, extracting the full frequency band energy data of the anomaly signal segment of a mold microcrack, and after normalization processing, yields a standardized set of frequency domain energy distributions.

[0046] Next, the frequency domain energy distribution data is integrated in chronological order to form a time-series feature sequence. The integration operation strictly follows the sampling sequence of the original signal, concatenating the frequency domain data from each time slice according to the sampling order. The sequence dimension perfectly matches the original signal segments, and the timing alignment accuracy is maintained at the single-sample-point level, eliminating data misalignment or missing data. For example, concatenating all frequency domain energy data in millisecond-level sampling order generates a time-series feature sequence perfectly aligned with the time domain signal.

[0047] It should be noted that the energy threshold is set based on the energy statistics of abnormal signals from stamping equipment over the past year. The specific process for setting the preset energy threshold is as follows: 10,000 sets of signal energy statistics samples containing normal and known abnormal operating conditions are extracted from the historical operation database of stamping equipment over the past year; a dynamic dividing line is set within the normalized energy range of 0 to 1 with a step size of 0.01; the false negative rate of abnormal signals and the false positive rate of normal signals are calculated at different dividing points; a comprehensive evaluation function (such as an F1 score evaluation function) is constructed, incorporating the false negative rate and the false positive rate; and the cross-extreme point that makes the comprehensive evaluation function reach its global maximum value is found. Through massive data fitting calculations, when the dividing point is 0.6, the system's anomaly identification precision and recall reach the optimal balance. Therefore, the basic threshold is set to 0.6.

[0048] To accommodate the varying fault tolerance levels across different production processes, the threshold can be lowered to 0.5 for high-sensitivity fault monitoring scenarios (such as precision electronic component stamping, where the risk of missed alarms needs to be minimized), and raised to 0.7 for routine operating condition monitoring scenarios (such as thick plate processing with significant environmental vibration interference, where the false alarm rate needs to be controlled). Those skilled in the art can adjust this threshold within the range of 0.5 to 0.7 (inclusive) according to anomaly identification requirements.

[0049] By filtering out marker points in the time-series feature sequence that exceed a preset energy threshold, the time-domain interval and frequency band range in which the marker points appear consecutively are identified as abnormal energy accumulation areas, thus accurately locating the concentrated distribution range of abnormal energy. For example, consecutive sampling points with energy values ​​exceeding 0.6 are selected, and the time-domain interval from 120 milliseconds to 180 milliseconds is identified as the abnormal energy accumulation area.

[0050] Subsequently, the start and end positions of the abnormal energy accumulation region in the original signal are calculated to determine the abnormal distribution interval. The position conversion is based on the sampling point sequence number of the original signal, transforming it into uniform millisecond-level time-domain coordinates. The interval boundaries precisely match the signal sampling timing, and the finally determined abnormal distribution interval contains complete time-domain range and frequency band information, which can be directly used for subsequent fault evolution analysis. For example, converting the sampling point coordinates of the abnormal accumulation region determines the complete abnormal distribution interval from 115 milliseconds to 185 milliseconds in the original signal.

[0051] In step S106, extracting the signal feature sequence within the abnormal distribution interval and making predictions based on the signal feature sequence to obtain the predicted trend slope and predicted fluctuation amplitude includes: Extract the signal feature sequence within the abnormal distribution interval, the signal feature sequence including amplitude, frequency and phase information; Calculate the Euclidean distance between the signal feature sequence and the corresponding features of the preset historical baseline data to obtain the deviation value sequence; A prediction time window is set, and linear regression is performed on the data of the deviation value sequence within the prediction time window to obtain the predicted trend slope and the predicted fluctuation amplitude. The predicted trend slope and the predicted fluctuation amplitude are compared with a preset abnormal evolution pattern to obtain a similarity. If the similarity exceeds a preset evolution matching threshold, the predicted trend slope and the predicted fluctuation amplitude are output.

[0052] It should be noted that, firstly, signal feature sequences within the abnormal distribution interval are extracted. These sequences contain amplitude, frequency, and phase information. Feature extraction employs short-time Fourier transform (SFT) technology to separate three key features from the purified signal within the abnormal interval. All feature values ​​are mapped to the 0-1 range using a min-max normalization method, maintaining consistency with previous data. For example, from the abnormal distribution interval corresponding to microcracks in a mold, amplitude, frequency, and phase features are extracted, and after standardization, a complete signal feature sequence is obtained.

[0053] Next, the Euclidean distance between the signal feature sequence and the corresponding features of the preset historical baseline data is calculated to obtain the deviation value sequence. The preset historical baseline data is taken from the stable operation of the stamping equipment under no-load and interference-free conditions. The Euclidean distance is used to quantify the degree of deviation of the current feature from the normal state, and the calculation result is directly used as the deviation value sequence. For example, the Euclidean distance between the abnormal feature sequence and the standard baseline data is calculated point by point to generate a continuous deviation value sequence, which intuitively reflects the degree of signal deviation from the normal state.

[0054] Subsequently, a prediction time window was set, and linear regression was performed on the deviation value sequence within the prediction time window to obtain the predicted trend slope and predicted fluctuation amplitude. The prediction time window was set based on the temporal pattern of stamping equipment failure evolution over the past year, with a basic window length of 20 milliseconds. This could be shortened to 15 milliseconds for high-speed stamping scenarios and extended to 25 milliseconds for low-speed stamping scenarios. The least squares method was used for the fitting operation. The predicted trend slope characterizes the rate of change of the deviation, and the predicted fluctuation amplitude characterizes the dispersion of the deviation.

[0055] For example, by setting a prediction time window of 20 milliseconds and performing linear regression fitting on the deviation value sequence, a predicted trend slope of 0.25 and a predicted fluctuation amplitude of 0.3 are obtained.

[0056] It is worth noting that the evolution matching threshold is set based on statistical data of matching abnormal evolution patterns of stamping equipment over the past year. The minimum similarity value of all accurately matched faults is statistically analyzed, and the basic threshold is set to 0.7. For high-precision fault diagnosis scenarios, this threshold can be increased to 0.75, and for routine fault monitoring scenarios, it can be decreased to 0.65. This threshold has been verified in multiple scenarios and can accurately distinguish between valid fault predictions and invalid interference. The predicted trend slope and predicted fluctuation amplitude are compared with the preset abnormal evolution patterns to obtain the similarity. If the similarity exceeds the preset evolution matching threshold, the predicted trend slope and predicted fluctuation amplitude are output.

[0057] For example, if the calculated similarity between the predicted feature and the mold crack evolution pattern is 0.82, exceeding the basic threshold of 0.7, the system outputs the corresponding predicted trend slope and predicted fluctuation amplitude. In step S107, the step of performing polynomial regression fitting based on the predicted trend slope, predicted fluctuation amplitude, load data, and component wear data to obtain the fault evolution trajectory includes: Using the load data and component wear data as correction factors, the predicted trend slope and the predicted fluctuation amplitude are corrected to obtain the corrected deviation evolution data; The corrected deviation evolution data are fitted using multinomial regression to obtain a preliminary fault evolution path. The error variance is obtained by calculating the sum of squared residuals between the initial fault evolution path and the corrected deviation evolution data; The prediction confidence level is calculated based on the error variance. If the prediction confidence level exceeds a preset confidence threshold, the preliminary fault evolution path is determined as the fault evolution trajectory.

[0058] It should be noted that, firstly, load data and component wear data are used as correction factors to adjust the predicted trend slope and predicted fluctuation amplitude, resulting in corrected deviation evolution data. The correction weights are derived by weighted summation of the two types of factors, with load data weighted at 0.6 and component wear data weighted at 0.4. This weight combination is based on historical statistical data of stamping equipment failure evolution. Load changes directly alter the failure development rate and have a more significant impact on deviation evolution, thus receiving a higher weight. Component wear, being a long-term cumulative factor, has a weaker effect and is therefore given a lower weight. All corrected data are mapped to the 0-1 range using the min-max normalization method to ensure data dimensionality consistency.

[0059] For example, by combining the load value of 0.8 and the wear value of 0.3, the predicted trend slope is weighted and corrected to obtain deviation evolution data that fits the actual working conditions.

[0060] Next, the corrected deviation evolution data is fitted using a third-order polynomial regression algorithm to obtain a preliminary fault evolution path. The regression fitting employs the least squares method for optimization. The training set is constructed from time-series fault deviation data of stamping equipment under different operating conditions over the past two years, covering samples from all scenarios including light and heavy loads and different wear levels. No additional activation function is involved in the fitting process. The path data is output after the fitting residuals stabilize, fully characterizing the nonlinear variation of the deviation over time. For example, performing third-order polynomial fitting on the corrected time-series deviation data generates continuously changing curve data, serving as the preliminary evolution path of equipment faults.

[0061] Subsequently, the sum of squared residuals between the initial fault evolution path and the corrected deviation evolution data is calculated to obtain the error variance. The residual calculation compares the difference between the fitted value and the measured value point by point. After squaring all the differences, they are summed, and the mean value is obtained by combining it with the total sample size to obtain the error variance. The smaller the variance value, the higher the fit between the fitted result and the measured data, and the data does not need to be normalized again.

[0062] For example, the difference between the fitted path and the measured deviation is calculated point by point, and the mean value is obtained after squaring and summing the results to obtain the error variance value that characterizes the fitting accuracy.

[0063] In this implementation case, the specific process for setting the preset confidence threshold is as follows: 10,000 sets of fitting verification samples containing known real fault evolution results are extracted from the historical operation database of stamping equipment over the past year; a dynamic dividing line is set within a confidence interval of 0.5 to 1.0 with a step size of 0.01; the path prediction accuracy (i.e., the proportion of samples whose fitted paths match the actual evolution) and the actual fault false alarm rate are calculated at different dividing points; a comprehensive evaluation function (such as the Youden index or F1 score) is constructed, incorporating the prediction accuracy and the false alarm rate, and the cross-extreme point that makes the comprehensive evaluation function reach its global maximum is found. Through massive data fitting calculations, when the dividing point is 0.85, the system achieves the optimal balance between eliminating interference from false evolution paths and retaining effective early warnings. Therefore, the basic threshold is set to 0.85.

[0064] To address the varying fault tolerance requirements of different production processes, the threshold can be increased to 0.9 for high-precision stamping scenarios (which demand extremely high diagnostic accuracy and require strict control over maintenance false alarm costs), and decreased to 0.8 for conventional production scenarios (which prioritize comprehensive fault interception and allow for a moderate reduction in confidence requirements). Those skilled in the art can adjust this threshold within the range of 0.8 to 0.9 (inclusive) based on diagnostic accuracy requirements.

[0065] If the calculated prediction confidence level is greater than or equal to the preset confidence threshold, the preliminary fault evolution path is determined as the final fault evolution trajectory. For example, if the calculated prediction confidence level is 0.88, which exceeds the basic threshold of 0.85, the fitted path is determined to be valid and is determined as the equipment fault evolution trajectory.

[0066] In summary, this invention discloses a fault diagnosis method for a stamping safety and pressure-resistant device. The method includes collecting multi-source data on device operating status, load, and component wear; obtaining a purified signal through multi-scale wavelet decomposition and threshold shrinkage denoising; extracting time-frequency joint features to construct a high-dimensional vector; determining anomalies through dual thresholds and matching historical fault patterns; locating abnormal energy accumulation intervals and predicting deviation trends and fluctuation amplitudes; and fusing load and wear data to perform polynomial regression to fit a high-confidence fault evolution trajectory. This method enables early and accurate fault diagnosis of stamping safety and pressure-resistant devices, meeting the core requirements of industrial automation operation and maintenance for production safety and process stability.

[0067] Reference Figure 2 The second embodiment of the present invention provides a fault diagnosis system for a stamping safety anti-pressure device, comprising: The data acquisition module is used to acquire the operating status data, load data, and component wear data of the stamping safety anti-pressure device; A signal denoising module is used to denoise the operating status data to obtain a denoised signal sequence. The feature extraction module is used to extract the time-domain statistical features and frequency-domain energy features of the denoised signal sequence, construct a high-dimensional feature vector, and analyze the fluctuation amplitude and trend slope of the high-dimensional feature vector under different time windows. An anomaly identification module is used to match the high-dimensional feature vector with a pre-acquired historical anomaly pattern to obtain an anomaly matching result if the fluctuation amplitude exceeds a preset amplitude judgment threshold and the trend slope exceeds a preset slope judgment threshold. The interval positioning module is used to perform frequency domain analysis based on the anomaly matching results to obtain a time-series feature sequence, locate the anomaly clustering region based on the time-series feature sequence, and determine the anomaly distribution interval. The trend prediction module is used to extract the signal feature sequence within the abnormal distribution interval and make predictions based on the signal feature sequence to obtain the predicted trend slope and predicted fluctuation amplitude. The trajectory fitting module is used to perform polynomial regression fitting based on the predicted trend slope, predicted fluctuation amplitude, load data, and component wear data to obtain the fault evolution trajectory.

[0068] It should be noted that the fault diagnosis system for a stamping safety anti-pressure device provided in this embodiment of the invention is used to execute all the process steps of the fault diagnosis method for a stamping safety anti-pressure device in the above embodiment. The working principles and beneficial effects of the two are one-to-one, so they will not be described again.

[0069] It should be noted that the system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the system embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0070] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A failure diagnosis method of a punch safety guard, characterized by, include: Acquire operating status data, load data, and component wear data of the stamping safety anti-pressure device; The operating status data is denoised to obtain a denoised signal sequence; Extract the time-domain statistical features and frequency-domain energy features of the denoised signal sequence, construct a high-dimensional feature vector, and analyze the fluctuation amplitude and trend slope of the high-dimensional feature vector under different time windows; If the fluctuation amplitude exceeds a preset amplitude judgment threshold and the trend slope exceeds a preset slope judgment threshold, then the high-dimensional feature vector is matched with the pre-acquired historical anomaly pattern to obtain an anomaly matching result. Based on the anomaly matching results, frequency domain analysis is performed to obtain a time-series feature sequence. Based on the time-series feature sequence, anomaly clustering regions are located, and anomaly distribution intervals are determined. Extract the signal feature sequence within the abnormal distribution interval, and make predictions based on the signal feature sequence to obtain the predicted trend slope and predicted fluctuation amplitude; The fault evolution trajectory is obtained by performing polynomial regression fitting based on the predicted trend slope, predicted fluctuation amplitude, load data, and component wear data.

2. The fault diagnosis method for the stamping safety anti-pressure device according to claim 1, characterized in that, The acquisition of operating status data, load data, and component wear data of the stamping safety and anti-pressure device includes: A sensor array is deployed to collect raw data on the operating status of the stamping safety and anti-pressure device. The raw data on the operating status includes vibration signals, pressure signals, and photoelectric sensing signals. The real-time load data of the stamping safety anti-pressure device is read through the equipment controller; The wear monitoring unit collects cumulative wear data of components. The original operating status data, the real-time load data, and the cumulative wear data of the components are respectively processed by format standardization to obtain operating status data, load data, and component wear data.

3. The fault diagnosis method for the stamping safety anti-pressure device according to claim 1, characterized in that, The step of denoising the operating status data to obtain a denoised signal sequence includes: Extract the fluctuation characteristics of the operating status data, and determine the level depth of the multi-scale decomposition based on the fluctuation characteristics; The running status data is decomposed based on the level depth to obtain high-frequency detail components and low-frequency approximation components; The high-frequency detail components are subjected to spectral analysis by fast Fourier transform to obtain the high-frequency band corresponding to the noise, and the total signal energy within the high-frequency band is calculated to obtain the noise energy value. If the noise energy value exceeds the preset noise judgment threshold, then the high-frequency detail component is subjected to threshold shrinkage processing to obtain the optimized high-frequency component; The optimized high-frequency components and the low-frequency approximation components are reconstructed to obtain a denoised signal sequence.

4. The fault diagnosis method for the stamping safety anti-pressure device according to claim 1, characterized in that, The process of extracting the time-domain statistical features and frequency-domain energy features of the denoised signal sequence, constructing a high-dimensional feature vector, and analyzing the fluctuation amplitude and trend slope of the high-dimensional feature vector under different time windows includes: A fixed-length sliding window is set to slice the denoised signal sequence to obtain multiple continuous signal segments; The mean, variance, peak factor, and kurtosis of each signal segment are calculated to obtain the time-domain statistical characteristics; Perform spectral analysis on each signal segment to extract the energy proportion, spectral peak value, and harmonic components of the characteristic frequency band, and obtain the frequency domain energy characteristics. The time-domain statistical features and the frequency-domain energy features are concatenated according to their dimensions to construct a high-dimensional feature vector; The Euclidean distance of the high-dimensional feature vectors corresponding to the continuous time window is calculated to obtain the fluctuation amplitude; By fitting the values ​​of each dimension of the high-dimensional feature vector across multiple consecutive time windows, the slope of the trend of each feature dimension changing over time can be obtained.

5. The fault diagnosis method for the stamping safety anti-pressure device according to claim 1, characterized in that, If the fluctuation amplitude exceeds a preset amplitude determination threshold and the trend slope exceeds a preset slope determination threshold, then the high-dimensional feature vector is matched with a pre-acquired historical anomaly pattern to obtain an anomaly matching result, including: The fluctuation amplitude is compared with a preset amplitude judgment threshold, and the trend slope is compared with a preset slope judgment threshold. If the comparison results all exceed the corresponding threshold, then the signal segment within the corresponding time window is extracted as the abnormal sequence to be detected. Calculate the difference between the feature distribution of the mutation sequence to be detected and the feature distribution of the historical abnormal patterns obtained in advance; If the difference is lower than the preset pattern matching threshold, the detected abnormal sequence is determined to match the historical abnormal pattern, and an abnormal matching result is obtained.

6. The fault diagnosis method for the stamping safety anti-pressure device according to claim 1, characterized in that, The step of performing frequency domain analysis based on the anomaly matching results to obtain a time-series feature sequence, and locating anomaly clustering regions and determining anomaly distribution intervals based on the time-series feature sequence includes: Based on the anomaly matching results, extract the frequency domain energy distribution data of the corresponding signal segment; The frequency domain energy distribution data are integrated in chronological order to form a temporal feature sequence; The marker points in the time-series feature sequence that exceed a preset energy threshold are selected, and the time-domain interval and frequency band range in which the marker points appear consecutively are identified as abnormal energy accumulation areas. Calculate the start and end positions of the abnormal energy accumulation region in the original signal to determine the abnormal distribution range.

7. The fault diagnosis method for the stamping safety anti-pressure device according to claim 1, characterized in that, The step of extracting the signal feature sequence within the abnormal distribution interval and making predictions based on the signal feature sequence to obtain the predicted trend slope and predicted fluctuation amplitude includes: Extract the signal feature sequence within the abnormal distribution interval, the signal feature sequence including amplitude, frequency and phase information; Calculate the Euclidean distance between the signal feature sequence and the corresponding features of the preset historical baseline data to obtain the deviation value sequence; A prediction time window is set, and linear regression is performed on the data of the deviation value sequence within the prediction time window to obtain the predicted trend slope and the predicted fluctuation amplitude. The predicted trend slope and the predicted fluctuation amplitude are compared with a preset abnormal evolution pattern to obtain the similarity. If the similarity exceeds a preset evolution matching threshold, the predicted trend slope and the predicted fluctuation amplitude are output.

8. The fault diagnosis method for the stamping safety anti-pressure device according to claim 1, characterized in that, The step of performing multinomial regression fitting based on the predicted trend slope, predicted fluctuation amplitude, load data, and component wear data to obtain the fault evolution trajectory includes: Using the load data and component wear data as correction factors, the predicted trend slope and the predicted fluctuation amplitude are corrected to obtain the corrected deviation evolution data; The corrected deviation evolution data are fitted using multinomial regression to obtain a preliminary fault evolution path. The error variance is obtained by calculating the sum of squared residuals between the initial fault evolution path and the corrected deviation evolution data; The prediction confidence level is calculated based on the error variance. If the prediction confidence level exceeds a preset confidence threshold, the preliminary fault evolution path is determined as the fault evolution trajectory.

9. A fault diagnosis system for a stamping safety anti-pressure device, characterized in that, include: The data acquisition module is used to acquire the operating status data, load data, and component wear data of the stamping safety anti-pressure device; A signal denoising module is used to denoise the operating status data to obtain a denoised signal sequence. The feature extraction module is used to extract the time-domain statistical features and frequency-domain energy features of the denoised signal sequence, construct a high-dimensional feature vector, and analyze the fluctuation amplitude and trend slope of the high-dimensional feature vector under different time windows. An anomaly identification module is used to match the high-dimensional feature vector with a pre-acquired historical anomaly pattern to obtain an anomaly matching result if the fluctuation amplitude exceeds a preset amplitude judgment threshold and the trend slope exceeds a preset slope judgment threshold. The interval positioning module is used to perform frequency domain analysis based on the anomaly matching results to obtain a time-series feature sequence, locate the anomaly clustering region based on the time-series feature sequence, and determine the anomaly distribution interval. The trend prediction module is used to extract the signal feature sequence within the abnormal distribution interval and make predictions based on the signal feature sequence to obtain the predicted trend slope and predicted fluctuation amplitude. The trajectory fitting module is used to perform polynomial regression fitting based on the predicted trend slope, predicted fluctuation amplitude, load data, and component wear data to obtain the fault evolution trajectory.

Images