Anti-vibration intelligent detection system and method for precision equipment

Abstract

The invention belongs to the technical field of intelligent detection, and provides an anti-vibration intelligent detection system and method for precision equipment, and the system comprises a data collection module, a data communication module and an intelligent processing module. The data acquisition module is used for acquiring vibration data and environment data in the operation environment of the precision equipment, and sending the data to the intelligent processing module through the data communication module by using an encryption channel; the intelligent processing module is used for carrying out space-time alignment on the vibration data and the environment data, constructing a fusion data set containing environment correlation features, establishing a statistical classification model and a deep learning model, respectively inputting the fusion data set into the statistical classification model and the deep learning model to obtain a first anomaly detection result and a second anomaly detection result, and outputting the first anomaly detection result and the second anomaly detection result; and fusing the first abnormal detection result and the second abnormal detection result to obtain a vibration detection result. According to the scheme, the accuracy, the high efficiency and the convenience of a vibration detection link are improved according to the vibration data and the environment data and in combination with various analysis models.

Description

Technical Field

[0001] This invention relates to the field of intelligent detection technology, and in particular to an intelligent vibration-damping detection system and method for precision equipment. Background Technology

[0002] In today's precision electronic components industry, the stable operation of precision equipment plays a decisive role in production quality and efficiency. However, vibration interference in the equipment's operating environment has become a key factor affecting its performance; therefore, vibration detection of precision equipment is particularly important.

[0003] In related technologies, traditional vibration detection methods mostly rely on regular manual inspections using simple handheld vibration detection instruments. This is not only inefficient but also prone to inaccurate results due to variations in human operation, making it difficult to achieve real-time and accurate vibration monitoring. Some intelligent detection solutions employ simple data statistics and threshold judgment methods, which are insufficient to extract deeper information from massive and complex vibration data, resulting in inaccurate judgments of vibration anomalies and failing to meet actual detection needs.

[0004] It is evident that traditional vibration detection methods suffer from low detection accuracy and are not efficient or convenient enough. Summary of the Invention

[0005] This invention provides an intelligent vibration detection system and method for precision equipment, which solves the shortcomings of traditional vibration detection schemes, such as low detection accuracy and lack of efficiency and convenience.

[0006] On the one hand, the present invention provides an intelligent vibration detection system for precision equipment, comprising: a data acquisition module, a data communication module, and an intelligent processing module; The data communication module is connected to the data acquisition module and the intelligent processing module respectively; The data acquisition module is used to collect vibration data and environmental data in the operating environment of precision equipment, and to send the vibration data and environmental data to the intelligent processing module through the data communication module using an encrypted channel. The intelligent processing module is used to perform spatiotemporal alignment of the vibration data and the environmental data, construct a fusion dataset containing environmental correlation features, and establish a statistical classification model and a deep learning model. The fusion dataset is input into the statistical classification model and the deep learning model respectively to obtain a first anomaly detection result and a second anomaly detection result. The first detection result and the second anomaly detection result are fused to obtain the vibration detection result.

[0007] According to the intelligent vibration detection system for precision equipment provided by the present invention, the environmental data includes: temperature and humidity data and air pressure data; The data acquisition module includes: Vibration sensors are used to collect vibration data during the operation of precision equipment; Temperature and humidity sensors are used to collect temperature and humidity data in the operating environment of precision equipment; A barometric pressure sensor is used to collect barometric pressure data in the operating environment of precision equipment.

[0008] According to the intelligent vibration detection system for precision equipment provided by the present invention, the data communication module includes: An industrial Ethernet communication submodule is used to send vibration data and environmental data collected by data acquisition modules deployed at surrounding fixed monitoring points to the intelligent processing module via an encrypted channel; The 5G communication submodule is used to send vibration data and environmental data collected by the data acquisition module deployed on the mobile monitoring device to the intelligent processing module via an encrypted channel; The LoRa communication submodule is used to send vibration data and environmental data collected by the data acquisition module deployed at remote fixed monitoring points to the intelligent processing module via an encrypted channel.

[0009] According to the intelligent vibration detection system for precision equipment provided by the present invention, the intelligent processing module performs spatiotemporal alignment of the vibration data and the environmental data to construct a fusion dataset containing environmental correlation features, including: Based on timestamp information, the vibration data and the environmental data are time-calibrated; Based on the physical coordinates of each sensor in the data acquisition module, spatial matching is performed on the time-calibrated vibration data and environmental data; The spatially matched vibration data and environmental data are denoised and normalized to obtain preprocessed data. Environmental association features are extracted from the preprocessed data to obtain a fused dataset containing environmental association features.

[0010] According to the intelligent vibration detection system for precision equipment provided by the present invention, the statistical classification model includes: The first preprocessing layer is used to perform feature optimization processing on the fused dataset to obtain the first input data; The SVM computation layer is used to perform classification calculations on the first input data based on a pre-defined kernel function and key parameters to obtain the first output result. The random forest computation layer is used to classify the first input data by voting using multiple pre-set decision trees to obtain the second output result; The XGBoost computation layer is used to classify and compute the first input data through a gradient boosting iterative optimization strategy to obtain a third output result. The fusion output layer is used to fuse the first output result, the second output result, and the third output result, and output the first anomaly detection result.

[0011] According to the intelligent vibration detection system for precision equipment provided by the present invention, the deep learning model includes: The second preprocessing layer is used to perform format conversion and data augmentation on the fused dataset to obtain the second input data; The CNN processing layer is used to extract the vibration local features from the second input data, and to perform dimensionality reduction and format conversion on the vibration local features to obtain a one-dimensional feature vector. The LSTM processing layer is used to perform temporal feature analysis on the first-dimensional feature vector to obtain the temporal feature matrix. The classification output layer is used to convert the temporal feature matrix into anomaly classification probabilities and output a second anomaly detection result.

[0012] According to the intelligent vibration detection system for precision equipment provided by the present invention, the CNN processing layer includes: Convolutional sublayers are used to extract local vibration features from the second input data through sliding calculations of the convolutional kernels; A pooling sublayer is used to reduce the dimensionality of the vibration local features using a max pooling strategy to obtain a two-dimensional feature map. The flattening sublayer is used to flatten the two-dimensional feature map into a one-dimensional initial vector and perform batch normalization on the one-dimensional initial vector to obtain a one-dimensional feature vector.

[0013] According to the intelligent vibration detection system for precision equipment provided by the present invention, the intelligent processing module is further configured to: input the fused dataset into a pre-built trend prediction model to obtain vibration trend prediction results; The trend prediction model is obtained by training a hybrid network containing a long short-term memory network and a gated recurrent network based on vibration samples, environmental samples, and trend prediction samples.

[0014] According to the intelligent vibration detection system for precision equipment provided by the present invention, the system further includes: an information interaction module; The information interaction module is connected to the intelligent processing module; The intelligent processing module is also used to generate and send abnormal warning information to the information interaction module when the vibration detection result is abnormal. The information interaction module is used to display vibration data, environmental data, vibration detection results, and abnormal warning information of precision equipment through a visual interface.

[0015] On the other hand, the present invention also provides an intelligent vibration detection method for precision equipment, which is implemented based on any of the above-described intelligent vibration detection systems for precision equipment, and the method includes: Vibration and environmental data are collected from the operating environment of precision equipment through a data acquisition module; The vibration data and environmental data are transmitted to the intelligent processing module via an encrypted channel through the data communication module. The vibration data and environmental data are spatiotemporally aligned by an intelligent processing module to construct a fusion dataset containing environmental correlation features. A statistical classification model and a deep learning model are established. The fusion dataset is input into the statistical classification model and the deep learning model respectively to obtain a first anomaly detection result and a second anomaly detection result. The first detection result and the second anomaly detection result are fused to obtain the vibration detection result.

[0016] The intelligent vibration detection system and method for precision equipment provided by this invention collects vibration data and environmental data from the operating environment of the precision equipment through a data acquisition module, and sends the vibration data and environmental data to an intelligent processing module via an encrypted channel through a data communication module. The intelligent processing module performs spatiotemporal alignment of the vibration data and environmental data, constructs a fusion dataset containing environmental correlation features, and establishes a statistical classification model and a deep learning model. The fusion dataset is input into the statistical classification model and the deep learning model respectively to obtain a first anomaly detection result and a second anomaly detection result. The first detection result and the second anomaly detection result are then fused to obtain the vibration detection result. This solution, based on the collected vibration data and environmental data and combined with multiple analysis models, can achieve automated detection of vibration throughout the entire process of precision equipment vibration, effectively improving the accuracy, efficiency, and convenience of the vibration detection process. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0018] Figure 1 This is a schematic diagram of the structure of the intelligent vibration detection system for precision equipment provided in an embodiment of the present invention; Figure 2 Yes, this is a schematic diagram of the hierarchical structure of a statistical classification model; Figure 3 This is a schematic diagram of the layer structure of a deep learning model; Figure 4 This is a flowchart illustrating the intelligent vibration detection method for precision equipment provided in this embodiment of the invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0020] The following is combined Figures 1 to 4 This invention describes the detailed scheme of the intelligent vibration detection system and method for precision equipment provided in the embodiments of the present invention.

[0021] like Figure 1 As shown, the intelligent vibration detection system for precision equipment provided in this embodiment of the invention mainly includes: a data acquisition module 110, a data communication module 120, and an intelligent processing module 130.

[0022] The data communication module 120 is connected to the data acquisition module 110 and the intelligent processing module 130 respectively.

[0023] The data acquisition module 110 is used to collect vibration data and environmental data in the operating environment of precision equipment, and sends the vibration data and environmental data to the intelligent processing module 130 through the data communication module 120 using an encrypted channel.

[0024] The intelligent processing module 130 is used to perform spatiotemporal alignment of vibration data and environmental data, construct a fusion dataset containing environmental correlation features, and establish a statistical classification model and a deep learning model. The fusion dataset is input into the statistical classification model and the deep learning model respectively to obtain the first anomaly detection result and the second anomaly detection result. The first detection result and the second anomaly detection result are fused to obtain the vibration detection result.

[0025] In this embodiment, the data acquisition module is the foundation of the entire detection system, primarily responsible for high-precision, multi-dimensional acquisition of vibration data and related environmental data from the operating environment of precision equipment. It not only enables real-time acquisition of vibration data but also allows for flexible adjustment of the acquisition frequency and range based on the vibration characteristics and detection requirements of different precision equipment. Simultaneously, it acquires environmental data for subsequent analysis of the correlation between vibration data and environmental factors, improving the accuracy of vibration impact assessment.

[0026] In one embodiment, the environmental data specifically includes temperature and humidity data and air pressure data.

[0027] Furthermore, such as Figure 1 As shown, the data acquisition module 110 specifically includes: Vibration sensor 1101 is used to collect vibration data during the operation of precision equipment.

[0028] Regarding the selection and deployment of vibration sensors, this embodiment comprehensively considers the vibration frequency range, sensitivity requirements, and application scenarios of different precision equipment. Specifically, piezoelectric, MEMS, and fiber optic vibration sensors can be selected and deployed in combination.

[0029] Among them, piezoelectric sensors are characterized by high sensitivity and a wide measurement frequency range, making them suitable for detecting high-frequency vibration signals, such as vibration data generated by the high-speed operation of precision machining equipment; MEMS sensors are characterized by small size, low power consumption, and low cost, and have good stability, making them suitable for large-scale deployment at multiple monitoring points in a factory to collect low- and medium-frequency vibration signals, such as vibration data caused by factory structure vibration and personnel movement; fiber optic sensors have the advantages of strong anti-electromagnetic interference capability, high measurement accuracy, and long transmission distance, and can be used in scenarios with extremely high vibration detection accuracy requirements, such as vibration monitoring of semiconductor manufacturing equipment.

[0030] In terms of deployment, a combination of fixed deployment in key areas and mobile flexible monitoring can be adopted. For key areas such as the vicinity of critical precision equipment and near the load-bearing columns of the factory, sensors can be fixedly installed with bolts to ensure the stability of the collected data. At the same time, portable mobile monitoring equipment can be equipped to collect vibration data of specific areas or equipment according to temporary detection needs.

[0031] Temperature and humidity sensor 1102 is used to collect temperature and humidity data in the operating environment of precision equipment.

[0032] The barometric pressure sensor 1103 is used to collect barometric pressure data in the operating environment of precision equipment.

[0033] In this embodiment, high-precision temperature and humidity sensors and barometric pressure sensors can be selected to collect data synchronously with the vibration sensor. Both the temperature and humidity sensors and the barometric pressure sensor have built-in data preprocessing chips, which can perform preliminary filtering and calibration on the collected temperature, humidity, and pressure data, reducing the impact of environmental interference on data accuracy. The temperature and humidity sensors and the barometric pressure sensor adopt an integrated acquisition node design with the vibration sensor, unified power supply and data output, ensuring the time synchronization of environmental data and vibration data, thereby providing a reliable data foundation for subsequent analysis of their correlation.

[0034] In this embodiment, the data communication module undertakes the important task of data transmission between the data acquisition module and the intelligent processing module. It needs to ensure the real-time performance, reliability, and security of data transmission. It not only needs to achieve efficient transmission of large amounts of vibration data and environmental data, but also needs to support bidirectional communication so that the control commands of the intelligent processing module can be accurately sent to the data acquisition module. At the same time, it has the ability to recover from network failures. When the communication link is interrupted, it can quickly switch to the backup link to avoid data loss.

[0035] In one embodiment, the data communication module specifically includes: The industrial Ethernet communication submodule is used to send vibration data and environmental data collected by the data acquisition module deployed at surrounding fixed monitoring points to the intelligent processing module via an encrypted channel.

[0036] The industrial Ethernet communication submodule serves as the core backbone of the system, undertaking high-frequency, high-capacity data transmission tasks from fixed monitoring points such as sensors on key precision equipment and load-bearing columns of the factory. The ring network redundancy of the industrial Ethernet communication submodule ensures uninterrupted core data transmission, while the segmented verification mechanism of the TCP / IP protocol ensures data accuracy, meeting the system's requirements for high reliability and low latency transmission to surrounding fixed monitoring points.

[0037] The 5G communication submodule is used to send vibration data and environmental data collected by the data acquisition module deployed on the mobile monitoring device to the intelligent processing module via an encrypted channel.

[0038] The 5G communication submodule serves as a high-speed, real-time supplementary link, primarily responsible for transmitting dynamic data from mobile monitoring devices such as portable vibration detectors. Since mobile monitoring devices cannot access fixed Ethernet networks, the wide coverage and low latency of 5G networks can meet the real-time data transmission requirements of these devices.

[0039] The LoRa communication submodule is used to send vibration and environmental data collected by the data acquisition module deployed at remote fixed monitoring points to the intelligent processing module via an encrypted channel.

[0040] The LoRa communication submodule, serving as a remote low-power link, is designed for remote monitoring points in factory buildings where cabling is difficult, such as sensors in factory corners or temporary structures. It can transmit periodic, small-capacity data. For example, these monitoring points collect vibration and environmental data every 30 seconds. LoRa's low-power characteristics reduce cabling and power supply costs, and its anti-interference capabilities ensure that data is not lost in remote areas. Data transmission is only triggered when the data volume accumulates to a threshold or an anomaly is detected, thereby reducing channel occupancy.

[0041] In one embodiment, the intelligent processing module performs spatiotemporal alignment of vibration data and environmental data to construct a fusion dataset containing environmentally relevant features, specifically including: The first step is to perform time calibration on the vibration data and environmental data based on the timestamp information.

[0042] In this embodiment, since the data acquisition module adopts an integrated acquisition node, it is powered uniformly and synchronizes the sampling time through the built-in clock chip. At the same time, it records the physical coordinates of each sensor deployment, such as the key equipment ID and the factory area number. The vibration data containing sensor location, sampling time, and vibration parameters, along with the environmental data containing corresponding location, synchronization timestamp, temperature and humidity value, and air pressure value, are packaged into a location-time-parameter associated data package according to a preset format.

[0043] Then, the intelligent processing module can perform time calibration on the vibration data and environmental data based on the timestamp information, thereby eliminating abnormal data with time deviations exceeding the threshold.

[0044] The second step involves spatial matching of the time-calibrated vibration data and environmental data based on the physical coordinates of each sensor in the data acquisition module.

[0045] In this embodiment, vibration data from the same monitoring point and the same time period can be bound to environmental data based on the physical coordinates of each sensor to achieve spatial matching.

[0046] The third step is to perform noise reduction and normalization on the spatially matched vibration data and environmental data to obtain preprocessed data.

[0047] In the denoising process, noise present in the vibration data, such as electromagnetic interference and sensor noise, can be addressed using a combination of wavelet transform and adaptive low-pass filtering. First, wavelet transform is used to decompose the vibration signal into multi-scale components, breaking it down into wavelet coefficients at different frequency scales. Based on the differences in wavelet coefficient characteristics between noise and the signal, a threshold is set to process the wavelet coefficients, suppressing those corresponding to noise. Then, an adaptive low-pass filtering algorithm is employed, automatically adjusting the filter cutoff frequency according to the signal's frequency characteristics to further filter out high-frequency noise. This approach more effectively preserves useful information in the vibration data, improving the denoising effect.

[0048] In the normalization process, due to the different sensitivities and measurement ranges of different sensors, there are differences in the dimensions and numerical ranges between the collected vibration data and environmental data. To facilitate subsequent feature extraction and model training, this embodiment uses the Z-score normalization method to normalize the vibration data and environmental data, converting the data into standard normal distribution data with a mean of 0 and a standard deviation of 1. For environmental data, the Min-Max normalization method can also be used to map the data to the [0,1] interval based on its physical meaning and value range.

[0049] The fourth step is to extract environmental association features from the preprocessed data to obtain a fused dataset containing environmental association features.

[0050] In this embodiment, the environmental correlation features include the rate of change of environmental parameters, the correlation coefficient between environment and vibration, and the label of abnormal environmental events. These features can more accurately quantify the impact of environmental factors on vibration and provide key basis for subsequent intelligent diagnosis.

[0051] In this embodiment, the anomaly detection process adopts a dual-track detection scheme. On the one hand, explicit anomalies are identified based on a statistical classification model to obtain the first anomaly detection result; on the other hand, implicit anomalies are captured based on a deep learning model to obtain the second anomaly detection result; finally, the first detection result and the second anomaly detection result are fused together according to the anomaly confidence level to obtain the vibration anomaly detection result.

[0052] In one embodiment, such as Figure 2 As shown, the statistical classification model specifically includes: The first preprocessing layer 210 is used to perform feature optimization processing on the fused dataset to obtain the first input data.

[0053] In this embodiment, the first preprocessing layer 210 serves as the data entry point for the statistical classification model, responsible for receiving the fused dataset and providing high-quality input for subsequent calculations.

[0054] Understandably, the fused dataset can be presented in the form of data vectors, for example, it can be represented as [vibration mean, impulse index, wavelet packet high-frequency band energy, temperature value, humidity value, air pressure value].

[0055] In the feature optimization process, differentiated processing logic can be adopted for different computation layers. Specifically, for the random forest computation layer and the XGBoost computation layer, key features can be screened through feature importance assessment. The random forest computation layer can calculate the contribution of each feature to the fault classification result based on the Gini coefficient. For example, the contribution of the peak value of the vibration signal to the equipment resonance fault is higher than that of the ambient humidity. The XGBoost computation layer quantifies the importance of features through the gain value. Both automatically remove redundant features with a contribution value lower than a preset threshold (such as 5%), such as irrelevant low-frequency environmental noise features, and retain 10-15 core features, such as vibration kurtosis, wavelet packet intermediate frequency energy, temperature deviation value, etc. For the SVM computation layer, feature distribution can be optimized through feature space mapping. Since the SVM computation layer relies on kernel functions to process linearly inseparable data, this layer maps the filtered low-dimensional features to a high-dimensional feature space through Gaussian kernel functions. This allows vibration fault data that were originally linearly inseparable, such as slight structural vibration and slight equipment imbalance, to be linearly separated in the high-dimensional space, laying the foundation for subsequent classification calculations.

[0056] The SVM computation layer 220 is used to perform classification calculations on the first input data based on a pre-defined kernel function and key parameters to obtain the first output result.

[0057] In this embodiment, the SVM computation layer 220 can perform classification calculations based on kernel functions and key parameters. First, the key parameters obtained by particle swarm optimization are loaded. For example, the Gaussian kernel function σ is 0.8 and the penalty factor C is 10. Then, the maximum margin hyperplane is found in the high-dimensional feature space to classify the first input data into different fault categories, such as normal, equipment resonance, and structural vibration. At the same time, the penalty factor C is used to balance the model's fit and generalization ability to the training data to avoid overfitting.

[0058] The random forest computation layer 230 is used to classify the first input data by voting using multiple pre-set decision trees to obtain the second output result.

[0059] In this embodiment, the random forest computing layer 230 achieves voting classification by integrating multiple decision trees. In practical applications, 100-200 independent decision trees can be constructed. Each decision tree is trained based on the training subset obtained by Bootstrap sampling. The first input data is used to make branch judgments. If the vibration peak value is greater than 0.5g, it enters the abnormal branch of the device; otherwise, it enters the normal branch. Finally, the classification results of all decision trees are counted, and the second output result is obtained according to the principle of majority rule.

[0060] XGBoost computation layer 240 is used to perform classification calculations on the first input data through a gradient boosting iterative optimization strategy to obtain the third output result.

[0061] In this embodiment, the XGBoost computation layer 240 achieves classification through a gradient boosting iterative optimization strategy. Specifically, the residual of the previous model is used as the optimization objective, and weak classifiers (i.e. decision trees) are built round by round. Each new decision tree focuses on correcting the diagnostic error of the previous round. For example, if a slightly resonant sample that was misclassified as normal in the previous round is identified by the new decision tree, the feature threshold is adjusted to strengthen the recognition of such samples. At the same time, a regularization term is introduced to suppress overfitting. Finally, the classification probability is output by weighted summation.

[0062] The fusion output layer 250 is used to fuse the first output result, the second output result, and the third output result, and output the first anomaly detection result.

[0063] Since the SVM computation layer can output clear fault category labels, the random forest computation layer can output category voting results, and the XGBoost computation layer can output probability distributions for each category, this embodiment uses a weighted voting mechanism to fuse the three output results. Specifically, weights can be assigned based on the model's historical diagnostic accuracy, and the output results of the three computation layers are weighted. For example, if both the XGBoost and random forest computation layers determine that it is structural vibration, the weighted score is 0.4 + 0.3 = 0.7, which is higher than the 0.3 score of the SVM computation layer for determining it as normal. Therefore, the first anomaly detection result is structural vibration.

[0064] In one embodiment, such as Figure 3 As shown, deep learning models specifically include: The second preprocessing layer 310 is used to perform format conversion and data augmentation on the fused dataset to obtain the second input data.

[0065] In this embodiment, the second preprocessing layer 310 can convert the fused dataset into a two-dimensional input format suitable for CNN, thereby laying the foundation for feature extraction.

[0066] Specifically, the second preprocessing layer 310 needs to complete two key processes: First, data format conversion, which slices the vibration data in the fusion dataset into 224×224 pixel vibration maps, where the horizontal axis represents time, the vertical axis represents amplitude, and the pixel value corresponds to the vibration intensity. At the same time, environmental data such as temperature, humidity, and air pressure are encoded into 1×N feature vectors and superimposed on the bottom of the vibration map to form a vibration-environment fusion input map. Second, data augmentation, which expands the dataset through operations such as random flipping, translation, and noise addition to avoid model overfitting.

[0067] The CNN processing layer 320 is used to extract the vibration local features from the second input data, and to perform dimensionality reduction and format conversion on the vibration local features to obtain a one-dimensional feature vector.

[0068] In a specific implementation, the CNN processing layer 320 specifically includes: Convolutional sublayer 3201 is used to extract local vibration features from the second input data through sliding calculation of the convolution kernel.

[0069] In this embodiment, the convolutional sub-layer 3201 adopts a three-layer convolutional structure. The first convolutional sub-layer uses 64 3×3 convolutional kernels with ReLU activation function, stride of 1, and padding of 1 pixel to ensure consistent input and output dimensions. It mainly extracts low-frequency global features of vibration data, such as the overall amplitude change trend and the basic correlation pattern between environmental parameters and vibration. For example, it captures the correlation feature of a slight increase in vibration amplitude when the temperature rises.

[0070] The second convolutional sublayer uses 128 3×3 convolutional kernels with ReLU activation function, stride of 1, and padding of 1 pixel. It focuses on extracting mid-frequency local features, such as the fundamental frequency component of precision equipment operation and periodic vibration pulses. For example, it can identify the second harmonic vibration signal generated by the wear of equipment bearings.

[0071] The third convolutional sublayer loads 256 pre-trained 3×3 convolutional kernel weights on a public mechanical fault dataset, freezes 60% of the bottom convolutional kernels to retain the ability to extract general fault features, and only fine-tunes 40% of the top convolutional kernels to adapt them to the specific vibration characteristics of precision equipment, such as the high-frequency micro-vibrations of semiconductor equipment. At the same time, an attention mechanism is introduced to strengthen the focus on fault-sensitive areas through the weight matrix, thereby improving the accuracy of feature extraction.

[0072] Pooling sublayer 3202 is used to reduce the dimensionality of local vibration features using a max pooling strategy to obtain a two-dimensional feature map.

[0073] In this embodiment, pooling sub-layers 3202 and convolutional sub-layers 3201 are alternately set, meaning that each convolutional sub-layer is followed by a pooling sub-layer. Specifically, the first pooling sub-layer has a 2×2 kernel size and a stride of 2, downsampling the low-frequency features output by the first convolutional sub-layer to retain global trend features and remove redundant environmental interference information. The second pooling sub-layer has a 2×2 kernel size and a stride of 2, focusing on retaining peak information in mid-frequency local features. The third pooling sub-layer has a 2×2 kernel size and a stride of 2, reducing the dimensionality of the high-frequency features after transfer learning fine-tuning to ensure that key fault features are not lost. Finally, the feature map corresponding to the 224×224 vibration local features is compressed into a 28×28 two-dimensional feature map, preparing for subsequent time-series analysis.

[0074] The flattening sublayer 3203 is used to flatten the two-dimensional feature map into a one-dimensional initial vector and perform batch normalization on the one-dimensional initial vector to obtain a one-dimensional feature vector.

[0075] In this embodiment, the flattening sublayer 3203 can flatten the two-dimensional feature map into a one-dimensional initial vector. At the same time, through batch normalization processing, the one-dimensional initial vector is standardized to a range with a mean of 0 and a standard deviation of 1 to obtain a one-dimensional feature vector, thus avoiding the influence of numerical differences in different feature dimensions on the temporal analysis accuracy of the LSTM processing layer.

[0076] The LSTM processing layer 330 is used to perform temporal feature analysis on the first-dimensional feature vector to obtain the temporal feature matrix.

[0077] In this embodiment, the LSTM processing layer 330 adopts a two-layer bidirectional LSTM structure. The first layer of the LSTM structure has 64 neurons, the activation function is tanh, the forget gate threshold is set to 0.8, and long-term temporal features are preferentially retained. It receives the one-dimensional feature vector output by the flattened layer, divides it into temporal sequences according to a set time step, and analyzes the changes in short-term vibration features.

[0078] The second-layer LSTM structure contains 128 neurons. The bidirectional LSTM structure captures future temporal trends forward and traces historical features backward with a dropout rate of 0.3. It focuses on long-term temporal dependency analysis. For example, it can identify the fault evolution trend where the vibration peak rises slowly over 30 minutes and has a lag correlation with a sudden drop in temperature. At the same time, it outputs the hidden state vector of each time step to form a temporal feature matrix.

[0079] The classification output layer 340 is used to convert the temporal feature matrix into anomaly classification probabilities and output the second anomaly detection result.

[0080] In this embodiment, the classification output layer 340 contains two fully connected layers. The first fully connected layer has 256 neurons, employs the ReLU activation function, and receives the temporal feature matrix output from the LSTM processing layer. Through linear transformation and nonlinear activation, it fuses local and temporal features. The second fully connected layer contains neurons matching the number of fault categories, with the Softmax activation function. It maps the fused feature vector to the probability distribution of each category and introduces an L2 regularization term to penalize excessively large weights, suppressing overfitting and ensuring the model has good generalization ability for unseen vibration patterns.

[0081] Finally, the category corresponding to the highest probability output in the two fully connected structure can be selected as the second anomaly detection result, and the probability value can be output. For example, the second anomaly detection result is that the fault category is structural loosening, and the probability value is 92%.

[0082] In one embodiment, the intelligent processing module can also be used to: input the fused dataset into a pre-built trend prediction model to obtain vibration trend prediction results.

[0083] The trend prediction model is obtained by training a hybrid network containing a long short-term memory network and a gated recurrent network based on vibration samples, environmental samples, and trend prediction samples.

[0084] In this embodiment, a hybrid network structure combining a Long Short-Term Memory (LSTM) network and a gated recurrent network is used to establish a trend prediction model. The LSTM network, a classic variant of the recurrent neural network, effectively handles time dependencies in long sequence data through forget gates, input gates, and output gates, and is particularly adept at capturing long-term trend features in vibration signals. The gated recurrent network, a lightweight improvement on LSTM, merges the forget gate and input gate into an update gate, while introducing a reset gate mechanism. This reduces the number of model parameters while preserving the ability to process sequence information, effectively improving training efficiency.

[0085] This embodiment deeply integrates Long Short-Term Memory (LSTM) networks and Gated Recurrent Networks (GRUs) through a parallel structure. The constructed trend prediction model can leverage the memory advantage of LSTM for complex vibration modes and the fast learning capability of GRUs to achieve a balance between prediction performance and computational efficiency. This enables it to accurately predict vibration change trends and provide effective data for subsequent decision-making regarding vibration anomalies.

[0086] In one embodiment, the above-mentioned intelligent vibration detection system for precision equipment may further include: an information interaction module, which is connected to the intelligent processing module.

[0087] The intelligent processing module is also used to generate and send abnormal warning information to the information interaction module when the vibration detection result indicates abnormal vibration.

[0088] The information interaction module is used to display vibration data, environmental data, vibration detection results, and abnormal warning information of precision equipment through a visual interface.

[0089] In this embodiment, the visualization interface can be divided into multiple areas. The left side displays the factory layout, showing the distribution of precision equipment and sensors within the factory in a 3D animation format. Different colored indicator lights indicate the working status of the sensors. The middle area displays real-time vibration waveforms and spectrum graphs, allowing users to zoom and pan to view waveform and spectrum details across different time periods and frequency ranges. Clicking on a point on the waveform graph displays the corresponding vibration data. The right side displays environmental data, presented in a dashboard format to intuitively reflect the current environmental conditions. Anomaly warnings can be marked on the factory layout, for example, by flashing bright spots on the layout based on the location of the anomaly.

[0090] Based on the same general inventive concept, this invention also protects an intelligent vibration detection method for precision equipment. The intelligent vibration detection method for precision equipment provided by this invention will be described below. The intelligent vibration detection method for precision equipment described below can be referred to in correspondence with the intelligent vibration detection system for precision equipment described above.

[0091] like Figure 4 As shown, the intelligent vibration detection method for precision equipment provided in this embodiment of the invention is based on the intelligent vibration detection system for precision equipment provided in the above embodiments. The method mainly includes the following steps: Step 410: Collect vibration data and environmental data in the operating environment of the precision equipment through the data acquisition module.

[0092] Step 420: Vibration data and environmental data are sent to the intelligent processing module via an encrypted channel through the data communication module.

[0093] Step 430: The vibration data and environmental data are spatiotemporally aligned using the intelligent processing module to construct a fusion dataset containing environmental correlation features. A statistical classification model and a deep learning model are established. The fusion dataset is input into the statistical classification model and the deep learning model respectively to obtain the first anomaly detection result and the second anomaly detection result. The first detection result and the second anomaly detection result are fused to obtain the vibration detection result.

[0094] Regarding the methods in the above embodiments, the specific implementation of each step has been described in detail in the embodiments of the relevant methods, and will not be elaborated further here.

[0095] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A vibration-proof intelligent detection system for precision equipment, characterized in that, The application relates to a precision equipment vibration detection system. The system comprises a data acquisition module, a data communication module and an intelligent processing module. The data communication module is connected to the data acquisition module and the intelligent processing module. The data acquisition module is used for collecting vibration data and environmental data in a precision equipment operating environment and sending the vibration data and the environmental data to the intelligent processing module through the data communication module by using an encrypted channel. The intelligent processing module is used for performing time-space alignment on the vibration data and the environmental data, constructing a fusion data set containing environmental correlation characteristics, establishing a statistical classification model and a deep learning model, inputting the fusion data set into the statistical classification model and the deep learning model respectively, obtaining a first abnormality detection result and a second abnormality detection result, fusing the first detection result and the second abnormality detection result, and obtaining a vibration detection result.

2. The anti-vibration intelligent detection system for precision equipment according to claim 1, characterized in that, The environmental data comprises temperature and humidity data and air pressure data. The data acquisition module comprises a vibration sensor, a temperature and humidity sensor and an air pressure sensor. The vibration sensor is used for collecting vibration data in a precision equipment operating process. The temperature and humidity sensor is used for collecting temperature and humidity data in a precision equipment operating environment. The air pressure sensor is used for collecting air pressure data in a precision equipment operating environment.

3. The anti-vibration intelligent detection system for precision equipment according to claim 1, characterized in that, The data communication module comprises an industrial Ethernet communication sub-module, a 5G communication sub-module and a LoRa communication sub-module. The industrial Ethernet communication sub-module is used for sending vibration data and environmental data collected by a data acquisition module arranged at a peripheral fixed monitoring point to the intelligent processing module by using an encrypted channel. The 5G communication sub-module is used for sending vibration data and environmental data collected by a data acquisition module arranged at a mobile monitoring device to the intelligent processing module by using an encrypted channel. The LoRa communication sub-module is used for sending vibration data and environmental data collected by a data acquisition module arranged at a remote fixed monitoring point to the intelligent processing module by using an encrypted channel.

4. The anti-vibration intelligent detection system for precision equipment according to claim 1, characterized in that, The intelligent processing module performs time-space alignment on the vibration data and the environmental data and constructs a fusion data set containing environmental correlation characteristics. The vibration data and the environmental data are time-calibrated based on timestamp information. The vibration data and the environmental data after time calibration are spatially matched based on the physical coordinates of the sensors in the data acquisition module. The vibration data and the environmental data after spatial matching are denoised and normalized to obtain preprocessed data. The preprocessed data are subjected to environmental correlation characteristic extraction to obtain a fusion data set containing environmental correlation characteristics.

5. The anti-vibration intelligent detection system for precision equipment according to claim 1, characterized in that, The statistical classification model comprises a first preprocessing layer, an SVM calculation layer, a random forest calculation layer and an XGBoost calculation layer. The first preprocessing layer is used for performing feature optimization processing on the fusion data set to obtain first input data. The SVM calculation layer is used for performing classification calculation on the first input data based on a pre-set kernel function and key parameters to obtain a first output result. The random forest calculation layer is used for performing voting classification on the first input data by using a pre-set plurality of decision trees to obtain a second output result. The XGBoost calculation layer is used for performing classification calculation on the first input data by using a gradient boosting iteration optimization strategy to obtain a third output result. A fusion output layer is configured to fuse the first output result, the second output result, and the third output result, and output a first anomaly detection result.

6. The anti-vibration intelligent detection system for precision equipment according to claim 1, characterized in that, The deep learning model comprises: A second pre-processing layer is configured to perform format conversion and data enhancement on the fusion dataset to obtain second input data; A CNN processing layer is configured to extract vibration local features in the second input data, and perform dimension reduction processing and format conversion on the vibration local features to obtain a one-dimensional feature vector; An LSTM processing layer is configured to perform time series feature analysis on the first-dimensional feature vector to obtain a time series feature matrix; A classification output layer is configured to convert the time series feature matrix into an anomaly classification probability, and output a second anomaly detection result.

7. The anti-vibration intelligent detection system for precision equipment according to claim 6, characterized in that, The CNN processing layer comprises: A convolutional sub-layer is configured to extract vibration local features in the second input data through convolution kernel sliding calculation; A pooling sub-layer is configured to perform dimension reduction processing on the vibration local features through a maximum pooling strategy to obtain a two-dimensional feature map; A flattening sub-layer is configured to flatten the two-dimensional feature map into a one-dimensional initial vector, and perform batch normalization processing on the one-dimensional initial vector to obtain a one-dimensional feature vector. 8.The anti-vibration intelligent detection system for precision equipment of claim 1, wherein, The intelligent processing module is further configured to input the fusion dataset into a pre-constructed trend prediction model to obtain a vibration trend prediction result. The trend prediction model is obtained by training a hybrid network comprising a long short-term memory network and a gated recurrent network based on vibration samples, environment samples, and trend prediction samples. 9.The anti-vibration intelligent detection system for precision equipment of claim 1, wherein, The system further comprises an information interaction module. The information interaction module is connected to the intelligent processing module. The intelligent processing module is further configured to generate and send an abnormal warning information to the information interaction module when the vibration detection result is a vibration anomaly. The information interaction module is configured to display the vibration data, the environment data, the vibration detection result, and the abnormal warning information of the precision equipment through a visual interface.

10. A method for intelligent anti-vibration detection of precision equipment, characterized in that, The method based on the anti-vibration intelligent detection system for precision equipment according to any one of claims 1 to 9 comprises: collecting vibration data and environment data in a precision equipment operating environment through a data acquisition module; sending the vibration data and the environment data to an intelligent processing module through an encrypted channel using a data communication module; performing spatio-temporal alignment on the vibration data and the environment data through an intelligent processing module, constructing a fusion dataset containing environment correlation features, and establishing a statistical classification model and a deep learning model, inputting the fusion dataset into the statistical classification model and the deep learning model respectively, obtaining a first anomaly detection result and a second anomaly detection result, and fusing the first detection result and the second anomaly detection result to obtain a vibration detection result.

Images