Shield Machine Operation Fault Prediction System and Method

Abstract

This application relates to the field of fault prediction technology, and discloses a shield machine operation fault prediction system and method. First, it obtains the amplitude data sequence of the shield machine cutter head, and adaptively adjusts the window length to adapt to the changes in actual working conditions, thereby improving the accuracy of fault prediction. By performing sequence comparison and analysis on the data of the initial and final windows, and using encoding and decoding technology to deeply mine the key information in the amplitude data, the accurate prediction of future amplitude values is realized. In this way, not only can potential faults be identified in a timely manner, but also a scientific basis is provided for preventive maintenance, significantly improving the safety and reliability of the shield machine operation.

Description

Technical Field

[0001] This application relates to the technical field of fault prediction, and more specifically, to a shield machine operation fault prediction system and method. Background Art

[0002] In modern underground engineering construction, as an important equipment for tunnel excavation, the operation status of a shield machine is directly related to the project progress and safety. However, with the increasing complexity of construction conditions, such as geological condition changes, groundwater pressure and temperature differences, etc., the cutterhead system of the shield machine faces severe challenges, which makes it particularly important to monitor the operation status of the shield machine in real time and predict faults. The traditional maintenance method of shield machines mainly relies on regular inspections and after-fact repairs. This method is not only inefficient, but also difficult to detect potential problems in advance, easily leading to sudden failures, causing huge economic losses and project delays.

[0003] At present, although some technologies have tried to conduct health assessments by monitoring the vibration of the shield machine cutterhead, these methods generally have certain limitations. For example, some solutions only focus on the trend analysis of a single parameter (such as amplitude), while ignoring the mutual influence between parameters under different working conditions; in addition, most of the existing algorithms use a fixed window length to process time series data, and this method seems not flexible enough when facing a dynamically changing working environment and cannot accurately capture early fault signals. In addition, traditional time series analysis means often lack the ability to extract deep features, resulting in limited accuracy of the prediction model and difficult to meet the requirements of practical applications.

[0004] Therefore, an optimized shield machine operation fault prediction solution is expected. Summary of the Invention

[0005] To solve the above technical problems, this application is proposed. Embodiments of this application provide a shield machine operation fault prediction system and method, which can not only identify potential faults in time, but also provide a scientific basis for preventive maintenance, significantly improving the safety and reliability of shield machine operation.

[0006] According to one aspect of the present application, a method for predicting operation faults of a shield machine is provided, including: obtaining an amplitude data sequence of the cutterhead of the shield machine; presetting an initial window length, and intercepting the amplitude data and rotational speed of the first initial window and the last initial window from the amplitude data sequence of the cutterhead of the shield machine to obtain the first initial window amplitude data and the last initial window amplitude data; based on the sequence comparison analysis between the first initial window amplitude data and the last initial window amplitude data, adjusting the initial window length to obtain a final window length; performing sliding window segmentation on the amplitude data sequence of the cutterhead of the shield machine based on the final window length to obtain a time series of amplitude window data; performing sequence encoding and decoding on the time series of amplitude window data to obtain an amplitude prediction value at a moment to be predicted; and generating the possibility of a fault of the shield machine at the moment to be predicted according to the difference between the amplitude prediction value at the moment to be predicted and the actual amplitude value at the moment to be predicted.

[0007] In the above method for predicting operation faults of a shield machine, based on the sequence comparison analysis between the first initial window amplitude data and the last initial window amplitude data, adjusting the initial window length to obtain a final window length includes: performing amplitude time series encoding on the first initial window amplitude data and the last initial window amplitude data to obtain an initial cutterhead amplitude time series encoding feature and a current cutterhead amplitude time series encoding feature; calculating a cutterhead amplitude time series difference feature between the initial cutterhead amplitude time series encoding feature and the current cutterhead amplitude time series encoding feature; decoding the cutterhead amplitude time series difference feature to obtain a window length adjustment coefficient; and multiplying the window length adjustment coefficient by the initial window length to obtain the final window length.

[0008] In the above method for predicting operation faults of a shield machine, calculating a cutterhead amplitude time series difference feature between the initial cutterhead amplitude time series encoding feature and the current cutterhead amplitude time series encoding feature includes: calculating the difference at each position between the initial cutterhead amplitude time series encoding feature and the current cutterhead amplitude time series encoding feature to obtain the cutterhead amplitude time series difference feature.

[0009] In the above method for predicting operation faults of a shield machine, performing sequence encoding and decoding on the time series of amplitude window data to obtain an amplitude prediction value at a moment to be predicted includes: performing amplitude time series encoding on each amplitude window data in the time series of amplitude window data to obtain a time series of amplitude local time series encoding features; performing time series context encoding on the time series of amplitude local time series encoding features to obtain an amplitude local time series pattern dynamic propagation encoding feature; and decoding the amplitude local time series pattern dynamic propagation encoding feature to obtain the amplitude prediction value at the moment to be predicted.

[0010] In the above-mentioned shield machine operation fault prediction method, performing temporal context encoding on the time series of the amplitude local temporal encoding features to obtain the amplitude local temporal pattern dynamic propagation encoding features includes: calculating the amplitude local end node association strength coefficient of each amplitude local temporal encoding feature in the time series of the amplitude local temporal encoding features; calculating the amplitude axial node membership factor of each amplitude local temporal encoding feature in the time series of the amplitude local temporal encoding features; calculating the spatio-temporal coupling constraint weight of each amplitude local temporal encoding feature based on the amplitude local end node association strength coefficient and the amplitude axial node membership factor of each amplitude local temporal encoding feature in the time series of the amplitude local temporal encoding features; and performing spatio-temporal coupling constraint propagation encoding on the time series of the amplitude local temporal encoding features based on the spatio-temporal coupling constraint weights of each amplitude local temporal encoding feature to obtain the amplitude local temporal pattern dynamic propagation encoding features.

[0011] In the above-mentioned shield machine operation fault prediction method, calculating the amplitude local end node association strength coefficient of each amplitude local temporal encoding feature in the time series of the amplitude local temporal encoding features includes: extracting the amplitude local temporal end encoding feature from the time series of the amplitude local temporal encoding features as the amplitude spatio-temporal propagation end anchor feature encoding vector; and calculating the amplitude local end node association strength coefficient of each amplitude local temporal encoding feature in the time series of the amplitude local temporal encoding features relative to the amplitude spatio-temporal propagation end anchor feature encoding vector.

[0012] In the above-mentioned shield machine operation fault prediction method, calculating the amplitude axial node membership factor of each amplitude local temporal encoding feature in the time series of the amplitude local temporal encoding features includes: performing clustering analysis on the time series of the amplitude local temporal encoding features to obtain the amplitude spatio-temporal propagation main axis reference feature encoding vector; and calculating the amplitude axial node membership factor of each amplitude local temporal encoding feature in the time series of the amplitude local temporal encoding features relative to the amplitude spatio-temporal propagation main axis reference feature encoding vector.

[0013] In the above shield machine operation fault prediction method, based on the amplitude local end node correlation strength coefficient and the amplitude axial node attribution factor of each amplitude local time series coding feature in the time series of the amplitude local time series coding features, calculate the spatio-temporal coupling constraint weight of each amplitude local time series coding feature, including: determining the axial propagation diffusion factor of each amplitude local time series coding feature; calculating the potential conflict specification factor of reverse propagation based on the amplitude local end node correlation strength coefficient and the amplitude axial node attribution factor; respectively performing axial propagation domain compact covariant modulation on the amplitude axial node attribution factor and the amplitude local end node correlation strength coefficient based on the potential conflict specification factor of reverse propagation and the axial propagation diffusion factor to obtain an optimized amplitude local end node correlation strength coefficient and an optimized amplitude axial node attribution factor; calculating the spatio-temporal coupling constraint weight of each amplitude local time series coding feature based on the optimized amplitude local end node correlation strength coefficient and the optimized amplitude axial node attribution factor.

[0014] In the above shield machine operation fault prediction method, decode the amplitude local time series pattern dynamic propagation coding feature to obtain the amplitude prediction value at the moment to be predicted, including: passing the amplitude local time series pattern dynamic propagation coding feature through an amplitude predictor based on a decoder to obtain the amplitude prediction value at the moment to be predicted.

[0015] According to another aspect of the present application, there is also provided a shield machine operation fault prediction system, including: an amplitude data sequence acquisition module for acquiring the amplitude data sequence of the cutter head of the shield machine; an initial window data extraction module for presetting an initial window length and intercepting the amplitude data and rotational speed of the first initial window and the last initial window from the amplitude data sequence of the cutter head of the shield machine to obtain the first initial window amplitude data and the last initial window amplitude data; a window length adjustment module for adjusting the initial window length based on the sequence comparison analysis between the first initial window amplitude data and the last initial window amplitude data to obtain a final window length; a sliding window segmentation module for performing sliding window segmentation on the amplitude data sequence of the cutter head of the shield machine based on the final window length to obtain a time series of amplitude window data; a sequence encoding and decoding module for performing sequence encoding and decoding on the time series of amplitude window data to obtain the amplitude prediction value at the moment to be predicted; a shield machine fault monitoring module for generating the possibility of shield machine fault at the moment to be predicted according to the difference between the amplitude prediction value at the moment to be predicted and the actual amplitude value at the moment to be predicted.

[0016] Compared with the prior art, the shield machine operation fault prediction system and method provided by the present application first obtain the amplitude data sequence of the shield machine cutter head, and adapt to the changes in actual working conditions by dynamically adjusting the window length, thereby improving the accuracy of fault prediction. By performing sequence comparison and analysis on the data of the initial and final windows, key information in the amplitude data is deeply mined using encoding and decoding techniques, and then accurate prediction of future amplitude values is achieved. In this way, not only can potential faults be identified in a timely manner, but also a scientific basis is provided for preventive maintenance, significantly enhancing the safety and reliability of the shield machine operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other objects, features, and advantages of the present application will become more obvious. The accompanying drawings are used to provide a further understanding of the embodiments of the present application, and constitute a part of the specification, and are used to explain the present application together with the embodiments of the present application, and do not constitute a limitation to the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0018] Figure 1 It is a schematic flowchart of the shield machine operation fault prediction method according to an embodiment of the present application.

[0019] Figure 2 It is a schematic diagram of data flow of the shield machine operation fault prediction method according to an embodiment of the present application.

[0020] Figure 3 It is a schematic flowchart of step 3 in the shield machine operation fault prediction method according to an embodiment of the present application.

[0021] Figure 4 It is a schematic flowchart of step 5 in the shield machine operation fault prediction method according to an embodiment of the present application.

[0022] Figure 5 It is a schematic flowchart of step 52 in the shield machine operation fault prediction method according to an embodiment of the present application.

[0023] Figure 6 It is a schematic block diagram of the shield machine operation fault prediction system according to an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0024] Hereinafter, exemplary embodiments of the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all of the embodiments of the present application. It should be understood that the present application is not limited by the exemplary embodiments described herein.

[0025] Figure 1 It is a schematic flowchart of the shield machine operation fault prediction method according to an embodiment of the present application. Figure 2Schematic diagram of data flow for the shield machine operation fault prediction method according to an embodiment of the present application. As Figure 1 and Figure 2 shown, the shield machine operation fault prediction method includes: Step 1: Obtain the amplitude data sequence of the shield machine cutter head; Step 2: Preset an initial window length, and intercept the amplitude data and rotational speed of the first initial window and the last initial window from the amplitude data sequence of the shield machine cutter head to obtain the first initial window amplitude data and the last initial window amplitude data; Step 3: Based on the sequence comparison analysis between the first initial window amplitude data and the last initial window amplitude data, adjust the initial window length to obtain the final window length; Step 4: Based on the final window length, perform sliding window segmentation on the amplitude data sequence of the shield machine cutter head to obtain the time series of amplitude window data; Step 5: Perform sequence encoding and decoding on the time series of amplitude window data to obtain the amplitude prediction value at the moment to be predicted; Step 6: Generate the possibility of shield machine failure at the moment to be predicted according to the difference between the amplitude prediction value at the moment to be predicted and the actual amplitude value at the moment to be predicted.

[0026] Specifically, in Step 1, obtain the amplitude data sequence of the shield machine cutter head. It should be understood that during the excavation process, the shield machine cutter head will be subjected to the reaction forces of surrounding soil, rock and other media, and this reaction force is directly reflected in the vibration mode of the cutter head. By monitoring the amplitude change of the cutter head, the change of the geological conditions ahead can be indirectly understood. For example, when passing through different types of soil layers or encountering obstacles, the vibration characteristics of the cutter head will change significantly. Therefore, collecting and analyzing these vibration data is crucial for predicting potential geological risks. In addition, the amplitude data can also reflect the working state of the internal mechanical components of the shield machine, such as whether there is abnormal wear or failure tendency.

[0027] In one embodiment, install high-precision vibration sensors on the shield machine cutter head, and these sensors can capture the subtle vibrations of the cutter head during the working process in real time. The selection of vibration sensors is crucial, and factors such as their sensitivity, response frequency range and anti-interference ability need to be considered to ensure the accuracy and reliability of the collected data. At the same time, in order to comprehensively reflect the working state of the cutter head, multiple sensors are usually arranged at different positions of the cutter head to form a comprehensive monitoring network. Once the sensors are deployed, the next is the data collection stage. During the operation of the shield machine, the vibration sensors continuously send signals to the data collection system. This process requires a stable and efficient communication link to transmit data to avoid any possible data loss or delay. Considering the particularity of the underground environment, wireless transmission solutions often fail to meet the requirements, so in most cases, wired connection methods are selected to ensure the stability and reliability of the data stream. The collected data will be stored in a central database for subsequent analysis.

[0028] In a specific embodiment, a shield machine equipped with an advanced vibration monitoring system is being used for construction of an underground tunnel project. The system includes a number of vibration sensors distributed on the cutter head. Each sensor collects vibration data of the cutter head at a set time interval (e.g., once per second) and sends it to the central control system. After receiving the data from each sensor, the central control system immediately preprocesses it, including removing background noise and high-frequency interference components, and then arranges the processed data in chronological order into a continuous amplitude data sequence. This sequence not only contains all the vibration information of the cutter head within a specific time period, but also reflects the dynamic response characteristics of the shield machine when passing through different types of geological structures.

[0029] Specifically, in step 2, a preset initial window length is set, and the amplitude data and rotational speed of the first initial window and the last initial window are intercepted from the amplitude data sequence of the cutter head of the shield machine to obtain the first initial window amplitude data and the last initial window amplitude data. It should be understood that the working performance of the shield machine under different geological conditions will be different, and these differences can be reflected by the vibration characteristics of the cutter head. However, due to the gradual and complex changes in geological conditions, the data at a single moment is not sufficient to comprehensively reflect the current and future working environment. Therefore, by setting an initial window length and selecting the data within a specific time period for analysis, the overall operation trend of the shield machine over a period of time can be better captured. In particular, the data of the first and last initial windows are selected because these two time periods usually represent the state when the shield machine starts to enter a new geological section and is about to leave the section, and the comparison between them can provide important information about geological changes. In addition, by combining the rotational speed data, the load condition and its stability of the cutter head under different geological conditions can be more accurately evaluated, providing a basis for subsequent adjustment of the window length.

[0030] In one embodiment, first, a reasonable initial window length is determined. The selection of this length needs to consider multiple factors, including the running speed of the shield machine, the rate of change of geological conditions, and data processing capabilities, etc. In a specific embodiment, the average propulsion speed of the shield machine is 1 meter per minute, and according to the previous exploration results, it is expected that the soil layer structure in front will change within about 50 meters. Then the initial window length can be set to 50 data points (i.e., corresponding to a time span of 50 seconds). This can not only ensure sufficient data volume to reflect the trend of geological changes but also avoid missing information on rapid changes due to an overly long window. Once the initial window length is determined, the next step is to intercept the data of the first and last initial windows from the entire amplitude data sequence. This step is usually completed by specialized data processing software. The software will automatically identify and extract the corresponding data segments according to predefined rules. For example, if the shield machine has been running for 300 seconds since startup and 50 seconds is selected as the window length according to the above settings, the software will extract the amplitude and rotational speed data from the data segment from the 1st second to the 50th second (the first initial window) and the data segment from the 251st second to the 300th second (the last initial window). These data not only contain the vibration intensity of the cutter head during the corresponding time period but also record the change in rotational speed, thus providing comprehensive basic data for subsequent analysis.

[0031] Specifically, in step 3, based on the sequence comparison analysis between the first initial window amplitude data and the last initial window amplitude data, the initial window length is adjusted to obtain the final window length. It should be understood that in actual operation, the preset initial window length may not be applicable to all cases because the rate and amplitude of geological condition changes may vary. For example, in some cases, the geological conditions change slowly and smoothly, and in this case, a longer window length may be more suitable; while in other cases, if the geological conditions change rapidly and violently, a shorter window length can better reflect such rapid changes. Therefore, by conducting a detailed comparison analysis of the data of the first initial window and the last initial window, the window length can be dynamically adjusted to adapt to different geological conditions.

[0032] In one embodiment, as Figure 3As shown, based on the sequence comparison analysis between the first initial window amplitude data and the last initial window amplitude data, adjusting the initial window length to obtain the final window length includes: Step 31: performing amplitude time series encoding on the first initial window amplitude data and the last initial window amplitude data to obtain the initial cutterhead amplitude time series encoding feature and the current cutterhead amplitude time series encoding feature; Step 32: calculating the cutterhead amplitude time series difference feature between the initial cutterhead amplitude time series encoding feature and the current cutterhead amplitude time series encoding feature; Step 33: decoding the cutterhead amplitude time series difference feature to obtain the window length adjustment coefficient; Step 34: multiplying the window length adjustment coefficient by the initial window length to obtain the final window length.

[0033] In Step 31, amplitude time series encoding is performed on the first initial window amplitude data and the last initial window amplitude data. This step aims to convert the original amplitude data into an easily processed form, namely the so-called amplitude time series encoding feature. In this way, the key patterns and trends in the data can be effectively extracted. In a specific embodiment, wavelet transform or Fourier transform can be used to encode the first initial window amplitude data and the last initial window amplitude data to generate the initial cutterhead amplitude time series encoding feature and the current cutterhead amplitude time series encoding feature.

[0034] In Step 32, the difference between these two amplitude time series encoding features is calculated. In one embodiment, calculating the cutterhead amplitude time series difference feature between the initial cutterhead amplitude time series encoding feature and the current cutterhead amplitude time series encoding feature includes: calculating the position-by-position difference between the initial cutterhead amplitude time series encoding feature and the current cutterhead amplitude time series encoding feature to obtain the cutterhead amplitude time series difference feature. Specifically, it is to compare the amplitude value differences at each corresponding time point within two time periods. This method is not only simple and direct but also can intuitively reflect the trend and amplitude of the amplitude change within two time periods. For example, if it is found that the amplitude fluctuation within the initial window is small while the amplitude fluctuation within the last window is large, it indicates that the shield machine may be entering a region with more complex geological conditions. In this way, the cutterhead amplitude time series difference feature can be obtained, providing a basis for subsequent window length adjustment.

[0035] In step 33, the decoding of the amplitude time series difference feature of the cutter head is performed to obtain the window length adjustment coefficient. The decoding process here is actually an inverse operation, that is, converting the difference feature into a specific value, which can be directly used to adjust the window length. It should be understood that if the amplitude difference feature obtained by calculation indicates that the current geological conditions change rapidly, the decoded window length adjustment coefficient may be smaller, meaning that the window length needs to be shortened to better capture this rapid change. On the contrary, if the geological conditions change relatively gently, the window length adjustment coefficient may be larger, allowing the use of a longer window length. In a specific embodiment, a pre-trained support vector regression (SVR) model is used for decoding. This model is trained based on a large amount of historical data from multiple similar projects and can effectively map the amplitude time series difference feature to the window length adjustment coefficient. When the amplitude time series difference feature of the cutter head is input, the SVR model outputs a window length adjustment coefficient of 0.8. Of course, the above is only an example and this embodiment is not specifically limited.

[0036] In step 34, multiplying the obtained window length adjustment coefficient by the initial window length can obtain the final window length. This process ensures that the window length can be dynamically adjusted according to the actual situation, thereby improving the accuracy of geological prediction. In a specific embodiment, the initially set window length is 50 seconds, and the window length adjustment coefficient obtained after the above series of analyses is 0.8, then the final window length will be adjusted to 40 seconds. The purpose of this is to make the window length more suitable for the current geological conditions, avoiding missing important information due to an overly long window or having overly fragmented data due to an overly short window, which affects the overall analysis effect.

[0037] Specifically, in step 4, based on the final window length, the amplitude data sequence of the shield machine cutter head is segmented by sliding window to obtain the time series of amplitude window data. It should be understood that through the above processing steps, a final window length for use is determined, and the amplitude data sequence is segmented by sliding window using this final window, which can better capture the overall operation trend of the shield machine over a period of time. This method can not only extract more detailed vibration information but also provide reliable data support for subsequent advanced analyses such as pattern recognition and fault prediction.

[0038] Specifically, in step 5, sequence encoding and decoding are performed on the time series of the amplitude window data to obtain the amplitude prediction value at the moment to be predicted. It should be understood that due to the complex and variable working environment of the shield machine, the data in a single time period is difficult to comprehensively reflect its long-term operation trend and potential risks. Therefore, by performing sequence encoding and decoding on the time series of the amplitude window data, more detailed vibration information can be extracted, and a model can be constructed to predict the amplitude value at a certain future moment. This method can not only improve the accuracy of geological forecasting but also enhance the equipment's ability to respond to emergencies.

[0039] In one embodiment, as Figure 4 shown, performing sequence encoding and decoding on the time series of the amplitude window data to obtain the amplitude prediction value at the moment to be predicted includes: Step 51: performing amplitude time series encoding on each amplitude window data in the time series of the amplitude window data to obtain a time series of amplitude local time series encoding features; Step 52: performing time series context encoding on the time series of the amplitude local time series encoding features to obtain amplitude local time series pattern dynamic propagation encoding features; Step 53: decoding the amplitude local time series pattern dynamic propagation encoding features to obtain the amplitude prediction value at the moment to be predicted.

[0040] In step 51, by performing amplitude time series encoding on each amplitude window data in the time series of the amplitude window data to obtain a time series of amplitude local time series encoding features, it is to extract more representative and discriminative feature representations from the original amplitude window data. This method can effectively capture the hidden patterns and trends within the time series data, which is crucial for accurately predicting the amplitude value at a future moment and identifying potential faults. The original amplitude window data may contain a large amount of noise and irrelevant information. Analyzing directly based on these data may lead to overfitting or underfitting of the model, thus affecting the prediction performance. Through the time series encoding step, the amplitude window data can be converted into a more compact and information-rich form, facilitating subsequent processing steps such as pattern recognition and anomaly detection tasks.

[0041] In a specific embodiment, a one-dimensional convolutional neural network (1D-CNN) can be used for amplitude time series encoding. First, each amplitude window data is regarded as an input sequence, and each sequence consists of a series of amplitude values arranged in chronological order. Then, an appropriate one-dimensional convolutional layer is designed, and the filter size is set according to actual needs. For example, 3 to 5 time steps can be selected as the width of the filter, which can reduce the computational complexity while retaining the features of the time dimension. For each amplitude window data, by applying this convolutional layer, multiple filters perform convolutional operations on the input sequence to automatically learn the local patterns and features in the input sequence. After each convolutional operation, an activation function such as ReLU is usually followed to increase non-linearity, which helps the model capture more complex patterns. Subsequently, a pooling layer such as max pooling can be used to further compress the features, reduce the number of parameters, and control overfitting. After several layers of such convolutional and pooling operations, a time series of amplitude local time series encoding features can be finally obtained.

[0042] In step 52, the process of performing temporal context encoding on the time series of amplitude local time series encoding features to obtain amplitude local time series pattern dynamic propagation encoding features is mainly aimed at more deeply mining the hidden patterns and trends inside the time series data, as well as the dynamic characteristics of these patterns changing over time. Specifically, after the amplitude window data undergoes preliminary amplitude time series encoding, although some basic feature representations can be extracted, these features are often limited to the local information within a single window and lack an understanding of the mutual relationships between different time periods in the entire time series. Through further temporal context encoding, the pattern evolution over a longer time span can be captured, thus better understanding the overall dynamic change law of the shield machine cutterhead vibration behavior. Moreover, in a complex working environment, changes in factors such as the geological conditions and operating parameters of the shield machine will cause corresponding changes in its vibration mode. It is difficult to comprehensively reflect the impact of these changes relying only on data at a single time point or in a short time window. Therefore, by performing temporal context encoding on the time series of amplitude local time series encoding features, the correlation between different moments can be comprehensively considered, potential trends and anomalies can be identified, and strong support can be provided for accurately predicting future amplitude values.

[0043] In one embodiment, as Figure 5As shown, performing temporal context encoding on the time series of the amplitude local temporal encoding features to obtain amplitude local temporal pattern dynamic propagation encoding features includes: Step 521: Calculating the amplitude local end node association strength coefficient of each amplitude local temporal encoding feature in the time series of the amplitude local temporal encoding features; Step 522: Calculating the amplitude axial node membership factor of each amplitude local temporal encoding feature in the time series of the amplitude local temporal encoding features; Step 523: Based on the amplitude local end node association strength coefficient and the amplitude axial node membership factor of each amplitude local temporal encoding feature in the time series of the amplitude local temporal encoding features, calculating the spatio-temporal coupling constraint weight of each amplitude local temporal encoding feature; Step 524: Based on the spatio-temporal coupling constraint weights of each amplitude local temporal encoding feature, performing spatio-temporal coupling constraint propagation encoding on the time series of the amplitude local temporal encoding features to obtain the amplitude local temporal pattern dynamic propagation encoding features.

[0044] In one embodiment, in Step 521, calculating the amplitude local end node association strength coefficient of each amplitude local temporal encoding feature in the time series of the amplitude local temporal encoding features includes: Extracting the amplitude local temporal end encoding feature from the time series of the amplitude local temporal encoding features as the amplitude spatio-temporal propagation end anchor feature encoding vector. Specifically, this process can be represented by the formula: ; where is the time series of the amplitude local temporal encoding features, and are respectively the 1st, 2nd, th, and th amplitude local temporal encoding features in the time series of the amplitude local temporal encoding features, and the th amplitude local temporal encoding feature is the amplitude local temporal end encoding feature, is the amplitude spatio-temporal propagation end anchor feature encoding vector.

[0045] Calculating the amplitude local end node association strength coefficient of each amplitude local temporal encoding feature in the time series of the amplitude local temporal encoding features relative to the amplitude spatio-temporal propagation end anchor feature encoding vector. Specifically, this process can be represented by the formula: ; where is the exponential function value with the natural constant as the base, is the feature value at the th position of , is the feature value at the th position of , is The number of eigenvalues is the correlation strength coefficient of the amplitude local end node corresponding to it.

[0046] It should be understood that the amplitude local time - series end - coding feature is extracted from the time series of the amplitude local time - series coding feature as the amplitude spatio - temporal propagation end - anchor feature coding vector, and the correlation strength coefficient of each amplitude local time - series coding feature relative to this end - anchor feature coding vector is calculated. This process aims to enhance the model's prediction ability for future trends by utilizing the latest information in the time series. Specifically, the last amplitude local time - series coding feature in the sequence, that is, the amplitude local time - series end - coding feature, is regarded as the representative of the current state or the sequence end. This approach provides a fixed reference point for the entire data - processing flow, enabling the system to consider the latest operating condition changes during the analysis. In this way, it can be ensured that the constructed model can not only capture the patterns in historical data but also dynamically adapt to new situations, improving the warning accuracy for potential faults. When calculating the correlation strength between each amplitude local time - series coding feature and the amplitude spatio - temporal propagation end - anchor feature coding vector, it actually quantifies the correlation or distance between the data at each time point and the current state. This step is similar to assigning a weight to each time point based on its similarity to the latest situation, thereby adjusting its importance in the overall analysis. For example, when identifying an impending mechanical fault, recent data is often more indicative than long - term data. Therefore, by introducing this end - constraint mechanism, it is possible to more precisely focus on those information segments that best reflect the current machine health state, helping to detect abnormal fluctuations in a timely manner and take preventive measures.

[0047] In one embodiment, in step 522, calculating the amplitude axial node membership degree factor of each amplitude local time - series coding feature in the time series of the amplitude local time - series coding feature includes: performing clustering analysis on the time series of the amplitude local time - series coding feature to obtain the amplitude spatio - temporal propagation main - axis reference feature coding vector. Specifically, this process can be represented by the formula: ; where, where, is the clustering analysis operation, and respectively take the maximum and minimum values of is the adjustment hyper - parameter. Among them, according to experience, the adjustment hyper - parameter is preset to 0.5, which can be adjusted and optimized according to the data situation. In this embodiment, it is not specifically limited. is the th amplitude local time - series coding feature reference anchor value in the sequence distribution of the amplitude local time - series coding feature constraint anchor value, is a normalization function, is the th amplitude local time-series encoding feature reference anchoring weight value in the time series of amplitude local time-series encoding feature constraint anchoring weight values, is the amplitude spatio-temporal propagation main axis reference feature encoding vector.

[0048] Calculate the amplitude axial node membership degree factor of each amplitude local time-series encoding feature in the time series of the amplitude local time-series encoding feature relative to the amplitude spatio-temporal propagation main axis reference feature encoding vector. Specifically, this process can be expressed by the formula: ; where, is the L2 norm of the calculation vector, is the inverse hyperbolic cosine function, represents corresponding amplitude axial node membership degree factor.

[0049] It should be understood that clustering analysis is performed on the time series of amplitude local time-series encoding features to extract the amplitude spatio-temporal propagation main axis reference feature encoding vector. This process aims to identify the main patterns or representative features in the dataset. These identified patterns can be regarded as the skeleton or the core components of the entire time series, which reflect the inherent basic structure and trend of the original data. In this way, not only can the most representative feature information be refined, but also a global reference framework can be provided for subsequent data processing steps. Further, calculating the amplitude axial node membership degree factor of each amplitude local time-series encoding feature relative to this amplitude spatio-temporal propagation main axis reference feature encoding vector is essentially to quantify the similarity or distance between the feature at each time point and the overall main axis. Such an operation assigns a weight value based on its relationship with the main axis to the feature at each time point, thereby adjusting the proportion of the importance of the feature in the entire analysis process. This helps to ensure that the information transmission not only conforms to the overall distribution trend of the data but also can accurately capture the changes in local details. For example, when monitoring the working state of a shield machine, this step can help the system more accurately identify behaviors that deviate from the normal working mode, even if these behaviors only account for a small proportion in the overall dataset. Therefore, by introducing this global constraint mechanism, not only can the generalization ability of the model be enhanced to prevent overfitting, but also the adaptability and stability of the model in the face of new working conditions can be improved, thereby improving the accuracy of fault prediction.

[0050] In one embodiment, based on the amplitude local end node association strength coefficient and the amplitude axial node membership degree factor of each amplitude local time-series encoding feature in the time series of the amplitude local time-series encoding feature, calculate the spatio-temporal coupling constraint weight of each amplitude local time-series encoding feature. Specifically, this process can be expressed by the formula: ; among them, and are respectively weighted hyperparameters, where the initial is set according to experience, and the weighted hyperparameters are trained with data and fine-tuned based on cross-validation techniques. is a step-type activation function, is the corresponding spatio-temporal coupling constraint weight.

[0051] It should be understood that the process of calculating the spatio-temporal coupling constraint weight based on the amplitude local end-node correlation strength coefficient and amplitude axial node membership factor of each feature in the time series of amplitude local time series coding features aims to assign a comprehensive weight value to each feature at each time point by integrating local and global information. This process not only considers the correlation of each feature with the current state (i.e., the end of the sequence), but also takes into account its position in the entire data structure and its consistency with the main trend. Specifically, by assigning a weight to each feature at each time point that combines local importance and global consistency, its true contribution in the overall analysis framework can be more accurately reflected. This fusion method can effectively balance the relationship between short-term fluctuations and long-term trends, so that the model will neither ignore the background trend due to over-focusing on the latest data, nor miss important immediate change signals due to over-reliance on historical patterns. For example, when analyzing the vibration data of the shield machine cutter head, this method can help the system better understand which vibration features are temporary anomalies and which are key indicators indicating potential failures. In this way, not only can the accuracy of equipment health condition assessment be improved, but also the ability of the model to warn of possible future failures can be enhanced.

[0052] In particular, for the sequence end feature conduction constraint of the amplitude local end-node correlation strength coefficient and the sequence transfer direction conduction constraint of the amplitude axial node membership factor, in order to be able to co-regulate the projection weights under the global structure constraint, it is necessary to consider the axial propagation domain diffusion interference of each original feature vector in the process of feature sequence information flow propagation.

[0053] In a preferred embodiment, in step 523, based on the amplitude local end-node correlation strength coefficient and amplitude axial node membership factor of each amplitude local time series coding feature in the time series of the amplitude local time series coding features, calculating the spatio-temporal coupling constraint weight of each amplitude local time series coding feature includes: first determining each amplitude local time series coding feature 's axial propagation diffusion factor, expressed as: ; among them, represents 's corresponding axial propagation diffusion gradient operator, represents The corresponding axial propagation diffusion factor.

[0054] And obtain the normalized regular fitting factor under the global constraint to coordinate the potential conflict between the global constraint limiting conditions and the normalized overflow risk. That is, based on the amplitude local end-node association strength coefficient and the amplitude axial node membership factor, calculate the cross-propagation potential conflict specification factor, expressed as: ; Indicating The corresponding cross-propagation potential conflict specification factor;

[0055] Next, based on the cross-propagation potential conflict specification factor and the axial propagation diffusion factor, perform axial propagation domain compact covariant modulation on the amplitude axial node membership factor and the amplitude local end-node association strength coefficient respectively to obtain the optimized amplitude local end-node association strength coefficient and the optimized amplitude axial node membership factor;

[0056] Specifically, for axial propagation domain compact covariant modulation, first use the axial propagation domain diffusion factor To perform axial propagation transformation on the amplitude axial node membership factor And apply the normalized regular fitting bias under the global constraint, expressed as: ; where Indicates the optimized amplitude axial node membership factor;

[0057] Then, through the normalized regular fitting factor under the global constraint, perform covariant transformation on the amplitude local end-node association strength coefficient based on the diffusion rotation transformation to impose a compact constraint: ; where Indicates the optimized amplitude local end-node association strength coefficient.

[0058] Finally, based on the optimized amplitude local end-node association strength coefficient and the optimized amplitude axial node membership factor, calculate the spatio-temporal coupling constraint weights of the respective amplitude local temporal coding features, expressed as: ; where And 9]Are weighted hyperparameters respectively, Is a step-type activation function, Is The corresponding spatio-temporal coupling constraint weight.

[0059] In this way, not only effectively balances the global structure of the feature sequence propagation and the local message passing dynamic characteristics, but also optimizes the amplitude local end-node association strength coefficient And the amplitude axial node membership factor Coordinate projection constraints within the sequence transfer field in the global structure, thus avoiding the implementation of axial transfer constraint coordination in the axial propagation domain diffusion interference that affects the tail message transfer.

[0060] In one embodiment, in step 524, based on the spatio-temporal coupling constraint weights of the respective amplitude local temporal coding features, spatio-temporal coupling constraint propagation coding is performed on the time series of the amplitude local temporal coding features to obtain the amplitude local temporal pattern dynamic propagation coding features. Specifically, this process can be expressed by the formula: ; where represents the amplitude local temporal pattern dynamic propagation coding features.

[0061] It should be understood that spatio-temporal coupling constraint propagation coding is performed on the time series of these features based on the spatio-temporal coupling constraint weights of the respective amplitude local temporal coding features to generate amplitude local temporal pattern dynamic propagation coding features. This process aims to refine a more accurate data representation by integrating the importance of features at each time point in their local and global contexts. Specifically, this approach first ensures that each feature not only reflects its correlation with the nearest state but also takes into account its position and trend consistency in the entire data structure. In this way, the hidden patterns and dynamic change laws within the data can be captured more accurately. When applying spatio-temporal coupling constraint weights for propagation coding, it is actually performing a dynamic weighted information transfer process. In this process, each feature is assigned a specific weight according to its relative importance from local and global perspectives, which helps to adjust its contribution degree in the final amplitude local temporal pattern dynamic propagation coding features.

[0062] In step 53, the amplitude prediction value at the moment to be predicted is obtained by decoding the amplitude local temporal pattern dynamic propagation coding features, and the high-level abstract features after deep processing and feature extraction are converted back into directly interpretable physical quantities - namely, the predicted amplitude value - for subsequent analysis and practical applications. Through the decoding step, the key information of the original signal can be recovered from the complex feature representation, and an accurate prediction of the amplitude at a future moment can be made based on this. This step can not only verify the effectiveness of feature extraction but also provide a direct basis for finally realizing fault prediction.

[0063] In one embodiment, decoding the amplitude local temporal pattern dynamic propagation coding feature to obtain the amplitude prediction value at the moment to be predicted includes: passing the amplitude local temporal pattern dynamic propagation coding feature through an amplitude predictor based on a decoder to obtain the amplitude prediction value at the moment to be predicted. Specifically, a one-dimensional convolutional neural network (1D CNN) can be used as part of the decoder to achieve the conversion from the amplitude local temporal pattern dynamic propagation coding feature to the amplitude prediction value. First, design a one-dimensional convolutional layer structure corresponding to the encoding stage, which is designed to reverse the operations performed during the encoding process, thereby gradually restoring the information of the original data. In this process, the encoded amplitude local temporal pattern dynamic propagation coding feature can be mapped back to an approximate representation of the original input space by using transposed convolution (deconvolution) or a combination of a series of carefully designed convolutional layers. For example, in the decoding stage, multiple one-dimensional convolutional layers can be utilized, each layer equipped with an appropriate filter size and stride to ensure that details at different scales can be captured. Meanwhile, to enhance the expressiveness of the model and facilitate the effective transmission of information, non-linear activation functions such as ReLU and batch normalization operations can be added after each layer of convolution. In addition, skip connections can be introduced to directly connect the outputs of some intermediate layers in the encoder to the corresponding layers in the decoder, thereby alleviating the vanishing gradient problem and helping to retain more original details. Finally, the output of the last layer of the decoder is mapped to a specific amplitude prediction value through a fully connected layer or a regression layer, completing the transformation from a complex feature representation to an interpretable physical quantity.

[0064] Specifically, in step 6, the possibility of shield machine failure at the moment to be predicted is generated according to the difference between the amplitude prediction value and the actual amplitude value at the moment to be predicted. It should be understood that by analyzing the difference between the amplitude prediction value and the actual value at the moment to be predicted, potential risk signals can be detected in a timely manner, and corresponding preventive measures can be taken. Specifically, if the cutter head vibrates normally next, its amplitude should be close to the predicted value; while if the amplitude of the cutter head differs significantly from the predicted value next, abnormal situations such as mechanical component failures of the shield machine may occur, and it is necessary to check the reasons for abnormal vibration in a timely manner and take measures. By this method, potential failure risks can be identified in advance, thus avoiding the occurrence of major accidents.

[0065] In one embodiment, the method for obtaining the possibility of shield machine failure at the moment to be predicted is: ; where, where, represents the possibility of shield machine failure at the moment to be predicted, represents the amplitude prediction value at the moment to be predicted, represents the actual amplitude value at the moment to be predicted, It represents a normalization function. The moment when the probability of shield machine failure is greater than the preset failure threshold is recorded as the moment when the shield machine fails. It should be noted that according to experience, the preset failure threshold is 0.5, which can be adjusted according to the actual situation and is not specifically limited in this embodiment.

[0066] In a preferred embodiment, the absolute difference between the predicted amplitude value at the moment to be predicted and the actual amplitude value at the moment to be predicted can be calculated by comparison. Then, the absolute difference and other relevant parameters (such as rotational speed, thrust, etc.) are input into a pre-trained support vector machine (SVM) classification model to generate the probability of shield machine failure at the moment to be predicted. For example, a probability value of 0.7 is output, indicating a relatively high probability of failure at that moment.

[0067] In summary, the shield machine operation fault prediction method provided in this application has been clarified. It first obtains the amplitude data sequence of the shield machine cutter head and adapts to the changes in actual working conditions by dynamically adjusting the window length, thereby improving the accuracy of fault prediction. Through sequence comparison and analysis of the data in the initial and final windows, key information in the amplitude data is deeply mined using encoding and decoding techniques, and then the accurate prediction of future amplitude values is realized. This method can not only identify potential faults in a timely manner but also provide a scientific basis for preventive maintenance, significantly improving the safety and reliability of shield machine operation.

[0068] This application also provides a shield machine operation fault prediction system, as Figure 6 shown. The shield machine operation fault prediction system 600 includes: an amplitude data sequence acquisition module 610 for acquiring the amplitude data sequence of the shield machine cutter head; an initial window data extraction module 620 for presetting an initial window length and intercepting the amplitude data and rotational speed of the first initial window and the last initial window from the amplitude data sequence of the shield machine cutter head to obtain the first initial window amplitude data and the last initial window amplitude data; a window length adjustment module 630 for adjusting the initial window length based on the sequence comparison and analysis between the first initial window amplitude data and the last initial window amplitude data to obtain the final window length; a sliding window segmentation module 640 for performing sliding window segmentation on the amplitude data sequence of the shield machine cutter head based on the final window length to obtain a time series of amplitude window data; a sequence encoding and decoding module 650 for performing sequence encoding and decoding on the time series of amplitude window data to obtain the predicted amplitude value at the moment to be predicted; and a shield machine fault monitoring module 660 for generating the probability of shield machine failure at the moment to be predicted according to the difference between the predicted amplitude value at the moment to be predicted and the actual amplitude value at the moment to be predicted.

[0069] An embodiment of the present application further provides a computer-readable storage medium. Computer program code is stored in the computer-readable storage medium. When the computer program code runs on a computer, the computer is enabled to execute the above-related method steps to implement an injection molded part detection method provided in the above embodiment.

[0070] An embodiment of the present application further provides a computer program product. When the computer program product runs on a computer, the computer is enabled to execute the above-related steps to implement an injection molded part detection method provided in the above embodiment.

[0071] Among them, the system, computer-readable storage medium, or computer program product provided in the embodiments of the present application are all used to execute the corresponding methods provided above. Therefore, the beneficial effects that can be achieved can refer to the beneficial effects in the corresponding methods provided above, and will not be elaborated here.

[0072] It should be noted that the above sequence of embodiments of the present application is only for description and does not represent the superiority or inferiority of the embodiments.

[0073] The processes depicted in the accompanying drawings do not necessarily require the particular order shown or sequential order to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous. Each embodiment in this specification is described in a progressive manner, and the same or similar parts among the embodiments can be referred to each other. The key point of each embodiment is to illustrate the differences from other embodiments.

Claims

1. A method for predicting operation faults of a shield machine, characterized in that Including: Obtain the amplitude data sequence of the shield machine cutter head; Preset an initial window length, and intercept the amplitude data and rotational speed of the first initial window and the last initial window from the amplitude data sequence of the shield machine cutter head to obtain the first initial window amplitude data and the last initial window amplitude data; Based on the sequence comparison analysis between the first initial window amplitude data and the last initial window amplitude data, adjust the initial window length to obtain the final window length; Based on the final window length, perform sliding window segmentation on the amplitude data sequence of the shield machine cutter head to obtain a time series of amplitude window data; Perform sequence encoding and decoding on the time series of the amplitude window data to obtain the amplitude prediction value at the moment to be predicted. Among them, the following temporal context encoding method is used to obtain the amplitude local temporal pattern dynamic propagation encoding feature, including: calculating the amplitude local end-node association strength coefficient of each amplitude local temporal encoding feature in the time series of the amplitude local temporal encoding feature; calculating the amplitude axial node membership factor of each amplitude local temporal encoding feature in the time series of the amplitude local temporal encoding feature; based on the amplitude local end-node association strength coefficient and the amplitude axial node membership factor of each amplitude local temporal encoding feature in the time series of the amplitude local temporal encoding feature, calculate the spatio-temporal coupling constraint weight of each amplitude local temporal encoding feature; based on the spatio-temporal coupling constraint weight of each amplitude local temporal encoding feature, perform spatio-temporal coupling constraint propagation encoding on the time series of the amplitude local temporal encoding feature to obtain the amplitude local temporal pattern dynamic propagation encoding feature. This process can be expressed by the formula: ; where is the th amplitude local temporal encoding feature in the time series of the amplitude local temporal encoding feature, and the th amplitude local temporal encoding feature is the amplitude local temporal end encoding feature, is the corresponding spatio-temporal coupling constraint weight, represents the amplitude local temporal pattern dynamic propagation encoding feature; Generate the possibility of shield machine failure at the moment to be predicted according to the difference between the predicted amplitude value and the actual amplitude value at the moment to be predicted.

2. The shield machine operation fault prediction method according to claim 1, characterized in that, Based on the sequence comparison analysis between the first initial window amplitude data and the last initial window amplitude data, adjusting the initial window length to obtain the final window length includes: Perform amplitude time series encoding on the first initial window amplitude data and the last initial window amplitude data to obtain the initial cutter head amplitude time series encoding feature and the current cutter head amplitude time series encoding feature; Calculate the cutter head amplitude time series difference feature between the initial cutter head amplitude time series encoding feature and the current cutter head amplitude time series encoding feature; Decode the cutter head amplitude time series difference feature to obtain the window length adjustment coefficient; Multiply the window length adjustment coefficient by the initial window length to obtain the final window length.

3. The shield machine operation fault prediction method according to claim 2, wherein Calculating the cutter head amplitude time series difference feature between the initial cutter head amplitude time series encoding feature and the current cutter head amplitude time series encoding feature includes: Calculate the position-by-position difference between the initial cutter head amplitude time series encoding feature and the current cutter head amplitude time series encoding feature to obtain the cutter head amplitude time series difference feature.

4. The shield machine operation fault prediction method according to claim 3, characterized in that, Performing sequence encoding and decoding on the time series of the amplitude window data to obtain the predicted amplitude value at the moment to be predicted includes: Perform amplitude time series encoding on each amplitude window data in the time series of the amplitude window data to obtain a time series of amplitude local time series encoding features; Perform time series context encoding on the time series of the amplitude local time series encoding features to obtain amplitude local time series pattern dynamic propagation encoding features; Decode the amplitude local time series pattern dynamic propagation encoding features to obtain the predicted amplitude value at the moment to be predicted.

5. The shield machine operation fault prediction method according to claim 4, wherein, Calculating the amplitude local end node association strength coefficient of each amplitude local time series encoding feature in the time series of the amplitude local time series encoding features includes: Extract the amplitude local time series end encoding feature from the time series of the amplitude local time series encoding features as the amplitude space-time propagation end anchor feature encoding vector; Calculate the amplitude local end node association strength coefficient of each amplitude local time series encoding feature in the time series of the amplitude local time series encoding features relative to the amplitude space-time propagation end anchor feature encoding vector.

6. The shield machine operation fault prediction method according to claim 5, wherein, Calculating the amplitude axial node membership factor of each amplitude local time series encoding feature in the time series of the amplitude local time series encoding features includes: Perform clustering analysis on the time series of the amplitude local temporal coding features to obtain the amplitude spatio-temporal propagation principal axis reference feature coding vector; Calculate the amplitude axial node membership degree factor of each amplitude local temporal coding feature in the time series of the amplitude local temporal coding features relative to the amplitude spatio-temporal propagation principal axis reference feature coding vector.

7. The shield machine operation fault prediction method according to claim 6, characterized in that, Based on the amplitude local end node correlation strength coefficient and the amplitude axial node membership degree factor of each amplitude local temporal coding feature in the time series of the amplitude local temporal coding features, calculate the spatio-temporal coupling constraint weight of each amplitude local temporal coding feature, including: Determine the axial propagation diffusion factor of each amplitude local temporal coding feature; Calculate the potential conflict specification factor for reverse propagation based on the amplitude local end node correlation strength coefficient and the amplitude axial node membership degree factor; Based on the potential conflict specification factor for reverse propagation and the axial propagation diffusion factor, perform axial propagation domain compact covariant modulation on the amplitude axial node membership degree factor and the amplitude local end node correlation strength coefficient respectively to obtain the optimized amplitude local end node correlation strength coefficient and the optimized amplitude axial node membership degree factor; Based on the optimized amplitude local end node correlation strength coefficient and the optimized amplitude axial node membership degree factor, calculate the spatio-temporal coupling constraint weight of each amplitude local temporal coding feature.

8. The shield machine operation fault prediction method according to claim 7, wherein Decode the amplitude local temporal pattern dynamic propagation coding feature to obtain the amplitude prediction value at the to-be-predicted moment, including: passing the amplitude local temporal pattern dynamic propagation coding feature through an amplitude predictor based on a decoder to obtain the amplitude prediction value at the to-be-predicted moment.

9. A shield machine operation fault prediction system for implementing the shield machine operation fault prediction method according to any one of claims 1-8, characterized in that, Including: An amplitude data sequence acquisition module for acquiring the amplitude data sequence of the shield machine cutter head; An initial window data extraction module for presetting an initial window length, and intercepting the amplitude data and rotational speed of the first initial window and the last initial window from the amplitude data sequence of the shield machine cutter head to obtain the first initial window amplitude data and the last initial window amplitude data; A window length adjustment module for adjusting the initial window length based on the sequence comparison analysis between the first initial window amplitude data and the last initial window amplitude data to obtain the final window length; A sliding window slicing module for performing sliding window slicing on the amplitude data sequence of the shield machine cutter head based on the final window length to obtain the time series of the amplitude window data; A sequence encoding and decoding module for performing sequence encoding and decoding on the time series of the amplitude window data to obtain the amplitude prediction value at the to-be-predicted moment; A shield machine fault monitoring module for generating the possibility of shield machine fault at the to-be-predicted moment according to the difference between the amplitude prediction value at the to-be-predicted moment and the actual amplitude value at the to-be-predicted moment.

Images