Rupture disc service life prediction method based on intelligent perception

Abstract

The invention relates to the technical field of industrial safety, and provides a rupture disk service life prediction method based on intelligent perception, which comprises the following steps: constructing a multi-mode intelligent sensor array, and collecting data in the running process of a rupture disk in real time; establishing a data transmission network, and transmitting data to a data processing center in real time; preprocessing the collected real-time data to form a standardized data set; and constructing a deep learning model, calculating the fatigue damage cumulant of the rupture disk according to the feature data output by the deep learning model based on a fatigue damage accumulation theory, predicting the residual life of the rupture disk, and generating life early warning information. According to the method, the running data of the rupture disk are collected in real time through the multi-mode intelligent sensor array, the life prediction model is constructed based on the fatigue damage theory, dynamic evaluation and early warning are achieved in combination with a three-level early warning mechanism, multi-source data are fused, the prediction precision is improved, the risk and cost are reduced, and it is guaranteed that industrial production is safe and reliable.

Description

Technical Field

[0001] The present invention relates to the field of industrial safety technology, and in particular to a bursting disc life prediction method based on intelligent perception. Background Art

[0002] A bursting disc is a safety pressure relief device consisting of a metal or non-metallic diaphragm and a retainer. When pressure within the equipment exceeds a predetermined value, the disc ruptures rapidly, releasing the pressure and preventing explosion or damage to the equipment due to overpressure. Characterized by sensitive operation, reliable sealing, and a simple structure, bursting discs are widely used in the chemical, petroleum, and power industries. As a critical safety pressure relief device, failure of a bursting disc can cause a major safety incident, while excessive replacement leads to increased maintenance costs.

[0003] Traditional life prediction methods rely heavily on empirical formulas or single sensor data, making it difficult to capture multi-dimensional failure characteristics under complex working conditions in real time, such as crack propagation, stress concentration, temperature and pressure fluctuations, and lack the ability to deeply analyze the spatiotemporal characteristics of the data. In existing technologies, problems such as irrational sensor layout leading to missed critical failure signals, insufficient data transmission security, noise interference affecting analysis accuracy, and the inability of simple model-based prediction methods to adapt to nonlinear fatigue damage processes are common, making it difficult to meet the needs of high-precision, real-time life prediction. Therefore, there is an urgent need for a prediction method that integrates multimodal perception, intelligent data processing, and deep learning to address the accuracy and reliability challenges of bursting disc life prediction in complex industrial environments. Summary of the Invention

[0004] In view of the shortcomings of the existing technology, the present invention provides a bursting disc life prediction method based on intelligent perception, which solves the problem of bursting disc life prediction in complex industrial environments that is difficult to solve in the existing technology.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: A method for predicting the life of a bursting disc based on intelligent perception, comprising the following steps:

[0006] S1. Construct a multimodal smart sensor array comprising an acoustic emission sensor, a strain sensor, a temperature sensor, and a pressure sensor. The multimodal smart sensor array is installed on the bursting disc and key surrounding locations according to a preset layout to collect real-time acoustic emission signals, strain information, temperature changes, and pressure fluctuation data during the operation of the bursting disc;

[0007] S2. Establish a data transmission network to transmit data collected by the multimodal intelligent sensor array to the data processing center in real time through a wireless transmission module, while encrypting the data;

[0008] S3. Preprocess the collected real-time data, using an adaptive filtering algorithm to remove noise based on the dynamic characteristics of the data. Standardization methods are used to unify data from different sensor types into a specific range to form a standardized data set.

[0009] S4. Construct a deep learning model consisting of a cascade of a convolutional neural network and a long short-term memory network. Input the standardized dataset into the deep learning model. The convolutional neural network extracts spatial features from the data, while the long short-term memory network analyzes the temporal characteristics of the data, mining the data for potential features related to the lifespan of the bursting disc.

[0010] S5. Based on the characteristic data output by the deep learning model and the fatigue damage accumulation theory, the formula Calculate the cumulative fatigue damage D of the bursting disc, where Δσ i is the stress amplitude of the i-th stress cycle, m is the material constant, C is a constant related to the material, and n is the number of stress cycles;

[0011] S6. Based on the accumulated fatigue damage of the bursting disc and the preset life threshold, the remaining life of the bursting disc is predicted. When the remaining life is lower than the set safety limit, a life warning message is generated.

[0012] Preferably, the preset layout is determined based on the structural characteristics and stress distribution of the bursting disc, with the acoustic emission sensor installed at a location where cracks are likely to form on the bursting disc, the strain sensor arranged in a stress concentration area, and the temperature sensor and pressure sensor installed on a medium flow path close to the bursting disc.

[0013] Preferably, in the step of establishing a data transmission network, the wireless transmission module adopts low-power wide area network technology to ensure stable data transmission in a complex industrial environment, and the data encryption processing adopts a combination of symmetric encryption and asymmetric encryption to improve the security of data transmission.

[0014] Preferably, the adaptive filtering algorithm is based on the minimum mean square error criterion and tracks noise changes in data in real time by continuously adjusting filter coefficients to achieve effective noise suppression.

[0015] Preferably, after the deep learning model is constructed, the historical failure data and normal operation data are used to train the deep learning model. During the training process, the stochastic gradient descent algorithm is combined with the momentum factor to adjust the model parameters to minimize the prediction error. The prediction error is calculated by the mean square error formula Calculate, where N is the number of samples, y j is the actual fatigue damage accumulation, is the fatigue damage accumulation predicted by the model.

[0016] Preferably, in the step of predicting the remaining life of the bursting disc, a mapping relationship model between the remaining life and the fatigue damage accumulation is established, and the remaining life of the bursting disc is determined by comparing the position of the current fatigue damage accumulation in the mapping relationship model. The mapping relationship model is an exponential decay model L 剩余 =L0·e -kD , where L0 is the initial life and k is the attenuation coefficient, which is determined by fitting historical data.

[0017] Preferably, the life warning information includes the remaining life value, warning level and replacement suggestion, which is sent to relevant operators through the industrial monitoring system in the form of sound and light alarms and SMS notifications. The warning level includes three threshold settings:

[0018] When the remaining life prediction value is lower than 80% of the design life, a yellow warning is triggered;

[0019] When the remaining life prediction value is lower than 50% of the design life, an orange warning is triggered;

[0020] When the remaining life prediction value is lower than 20% of the design life, a red warning is triggered and the equipment is shut down for protection.

[0021] Preferably, the multimodal intelligent sensor array is calibrated and maintained regularly, and the collected data is corrected according to the calibration results to ensure the accuracy and reliability of the data. At the same time, the training data of the deep learning model is updated to optimize the model prediction performance.

[0022] The present invention provides a method for predicting the life of a bursting disc based on intelligent perception. It has the following beneficial effects:

[0023] The present invention uses a multimodal intelligent sensor array to achieve real-time and accurate acquisition of acoustic emission signals, strain, temperature, and pressure data during the operation of the bursting disc. Combined with a deep learning model that combines a convolutional neural network with a long short-term memory network, it effectively extracts spatial and temporal features, deeply mines potential information related to fatigue damage, and constructs a remaining life prediction model based on the fatigue damage accumulation theory. Combining the exponential decay mapping relationship with a three-level warning mechanism, it achieves dynamic assessment and graded warning of the remaining life of the bursting disc. This method breaks through the limitations of traditional single-parameter monitoring and improves prediction accuracy through multi-source data fusion and intelligent algorithms. It can identify fatigue damage trends in advance, avoid the risk of sudden failure, and reduce excessive maintenance costs. It provides an intelligent solution for the safe operation of equipment in high-risk scenarios such as ferrous metal smelting, significantly improving the safety and reliability of industrial production. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 It is a perspective view of the present invention. DETAILED DESCRIPTION

[0025] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0026] Reference Figure 1 The embodiment of the present invention provides a method for predicting the life of a bursting disc based on intelligent perception, comprising the following steps:

[0027] S1. Construct a multimodal intelligent sensor array: Construct a multimodal intelligent sensor array that includes acoustic emission sensors, strain sensors, temperature sensors, and pressure sensors. Determine the preset layout based on the structural characteristics and stress distribution of the blasting disc: Install the acoustic emission sensor at the location where cracks are likely to form on the blasting disc to capture the acoustic emission signal when internal microcracks form and expand; arrange the strain sensor in the stress concentration area to monitor surface strain changes in real time; install the temperature sensor and pressure sensor on the medium flow path close to the blasting disc to obtain temperature and pressure data of the operating environment. Install each sensor in this layout on the blasting disc and surrounding key locations to collect real-time acoustic emission signals, strain information, temperature changes, and pressure fluctuation data during operation;

[0028] S2. Establish a data transmission network: A wireless transmission module using low-power wide area network technology transmits data collected by the multimodal intelligent sensor array to the data processing center in real time. To ensure data security, a combination of symmetric and asymmetric encryption is used to encrypt the data, preventing theft or tampering during transmission and ensuring stable data transmission in complex industrial environments.

[0029] S3. Data Preprocessing: Utilizing an adaptive filtering algorithm based on the minimum mean square error criterion, noise interference is removed based on the dynamic nature of the data. This algorithm continuously adjusts the filter coefficients to track noise changes in the data in real time, effectively suppressing noise. Subsequently, a normalization method is used to unify data from different sensor types into a specific range, forming a standardized dataset for subsequent analysis.

[0030] S4. Construct and train a deep learning model: Construct a deep learning model consisting of a convolutional neural network (CNN) and a long short-term memory network (LSTM) cascade, input a standardized data set into the model, and use CNN to extract spatial features: perform convolution operations through multiple convolution kernels of different sizes and parameters to extract features such as the local frequency of the acoustic emission signal and the local deformation of the strain data; after nonlinear activation, downsample through the pooling layer to reduce the data dimension and retain key features; finally, flatten the feature map into a one-dimensional vector to output the spatial features. LSTM analyzes the spatial features output by CNN in time series, controls the cell state update through the input gate, forget gate, and output gate, and captures the long-term dependency of the data, such as the cumulative change of stress over time. Finally, it outputs a hidden state vector to mine potential features related to the life of the blasting disc, such as the development trend of fatigue damage. Use historical failure data and normal operation data to train the model, and use the stochastic gradient descent algorithm combined with the momentum factor to adjust the parameters to minimize the mean square error. N is the number of samples, y j is the actual fatigue damage accumulation, Predict values for the model and optimize model performance;

[0031] S5. Calculate the cumulative amount of fatigue damage: Based on the characteristic data output by the deep learning model and the fatigue damage accumulation theory, use the formula Calculate the cumulative fatigue damage D of the bursting disc. Where, Δσ i is the stress amplitude of the i-th stress cycle, m is the material constant, C is a constant related to the material, and n is the number of stress cycles;

[0032] S6. Prediction of remaining life and early warning: Establish a mapping relationship model between remaining life and fatigue damage accumulation, using the exponential decay model L 剩余 =L0·e -kD , L0 is the initial life, and k is the attenuation coefficient, determined by fitting historical data. The remaining life is determined by comparing the current fatigue damage accumulation with the position in the model. Three levels of warning thresholds are set: when the predicted remaining life falls below 80% of the design life, a yellow warning is triggered; below 50%, an orange warning is triggered; and below 20%, a red warning is triggered, triggering equipment shutdown protection. The remaining life value, warning level, and replacement recommendations are transmitted to operators via audio and visual alarms and text message notifications via the industrial monitoring system.

[0033] Regularly calibrate and maintain the multimodal smart sensor array, and use the calibration results to correct the collected data to ensure accuracy and reliability. Simultaneously, update the training data of the deep learning model to further optimize the model's prediction performance and ensure long-term prediction accuracy.

[0034] In order to better illustrate the implementation scheme of the bursting disc life prediction method of the present invention, the following examples are given:

[0035] Example 1: Prediction of chemical reactor rupture disc life

[0036] A multimodal intelligent sensor array was constructed: An SR150M acoustic emission sensor with a frequency response range of 100kHz-400kHz was installed at the crown of the stainless steel dome-shaped bursting disc, accurately capturing acoustic emission signals generated by microcracks. A BX120-5AA resistance strain gauge with a measurement accuracy of 0.001με was attached to the stress concentration area at the edge of the bursting disc to monitor surface strain in real time. A Pt100 temperature sensor with a measurement range of -200°C-650°C and an accuracy of ±0.1°C and a CYB-201 pressure sensor with a range of 0-10MPa and an accuracy of 0.1%FS were installed on the media inlet pipe near the bursting disc to acquire real-time temperature and pressure data. Once these sensors were installed according to their layout, they collected real-time acoustic emission, strain, temperature, and pressure data at a frequency of 500Hz.

[0037] Establish a data transmission network: Use NB-IoT wireless transmission modules to transmit data to the data processing center in real time. The encryption method uses the AES-128 symmetric encryption algorithm to encrypt data and the RSA-2048 asymmetric encryption algorithm to transmit symmetric keys, ensuring stable and secure data transmission in complex industrial environments.

[0038] Data preprocessing: An adaptive filtering algorithm based on the minimum mean square error criterion was used, with a step size of 0.01, to adjust the filter coefficients in real time to suppress noise. Subsequently, the Z-score normalization method was used to convert the acoustic emission signal amplitude (mV), strain (με), temperature (°C), and pressure (MPa) to a standard normal distribution with a mean of 0 and a standard deviation of 1, forming a standardized data set.

[0039] Construct and train a deep learning model: The CNN part has two layers. The first layer uses 16 3×3 convolution kernels to extract features such as the local frequency of the acoustic emission signal. After ReLU activation, it is downsampled through a 2×2 maximum pooling layer. The second layer uses 32 5×5 convolution kernels to extract local deformation features of the strain data and flatten it into a one-dimensional vector. The LSTM part has 128 neurons, and the CNN output features are analyzed as a time series to capture long-term dependencies such as the cumulative changes in stress over time. The model was trained using 100 groups of historical failure data and 300 groups of normal operation data of the reactor bursting disc. The stochastic gradient descent algorithm was used with a learning rate of 0.001 and a momentum factor of 0.9 to adjust the parameters and minimize the mean square error. Optimize model performance.

[0040] Calculate the cumulative amount of fatigue damage: Based on the model output characteristic data, use the formula Calculation, where material constant m = 3.5, C = 1.2 × 10-11 , calculate the stress amplitude Δσ from the strain data i , combined with the number of stress cycles n recorded by the pressure sensor, the fatigue damage accumulation D is updated every 10 minutes.

[0041] Predicting Remaining Life and Early Warning: Establishing an Exponential Decay Model L 剩余 =L0·e -kD By fitting historical data, L0 = 8000 hours and k = 0.0008 were determined. When the predicted remaining life falls below 80% of the design life of 6000 hours, or 4800 hours, a yellow warning is triggered; below 50%, or 3000 hours, an orange warning is triggered; and below 20%, or 1200 hours, a red warning is triggered, leading to the reactor shutdown. The industrial monitoring system also notifies operators of the remaining life value, warning level, and replacement recommendations via audio and visual alarms and text messages.

[0042] Example 2: Life Prediction of Oil Pipeline Bursting Discs

[0043] A multimodal intelligent sensor array was constructed: a PAI-2000 acoustic emission sensor with a sensitivity of 70 dB was installed in the center of the carbon steel flat-plate bursting disc, a location prone to cracking, effectively capturing crack signals. BF120-3AA resistance strain gauges with a measurement accuracy of 0.002 με were attached to the stress concentration areas surrounding the bursting disc. A Cu50 temperature sensor with a measurement range of -50°C to 150°C and an accuracy of ±0.3°C and an MPM480 pressure sensor with a range of 0-16 MPa and an accuracy of 0.075% FS were installed in the pipeline medium flow path near the bursting disc. After installation, these sensors collected real-time data at a frequency of 300 Hz.

[0044] Establish a data transmission network: Use the LoRa wireless transmission module to transmit data to the data processing center. Encryption uses DES symmetric encryption with a 56-bit key length, combined with ECC asymmetric encryption with a 256-bit key length to ensure data security.

[0045] Data preprocessing: Adaptive filtering algorithm was used with an initial step size of 0.008 to remove noise. Then, the Min-Max normalization method was used to uniformly map the data from different sensors to the interval [0, 1] to form a standardized data set.

[0046] Constructing and training a deep learning model: The CNN consists of three layers. The first layer uses 32 3×3 convolution kernels, the second layer uses 64 5×5 convolution kernels, and the third layer uses 128 7×7 convolution kernels. Spatial features such as acoustic emission and strain are extracted and flattened using ReLU activation and 2×2 average pooling. The LSTM layer uses 256 neurons to analyze time series features. Training was performed using 80 sets of historical failure data and 200 sets of normal operation data from the pipeline bursting disc. Stochastic gradient descent was used with a learning rate of 0.001 and a momentum factor of 0.8 to adjust parameters and optimize the model.

[0047] Calculating fatigue damage accumulation: Using the formula Where m = 3.2, C = 8 × 10 -12 , according to the stress amplitude Δσ i The number of cycles n is calculated based on the pressure sensor data, and the fatigue damage accumulation D is calculated every 15 minutes.

[0048] Predicting remaining life and early warning: fitting exponential decay model L through historical data 剩余 =L0·e -kD , L0 = 6000 hours, k = 0.009. When the remaining life falls below 80% of the design life of 4000 hours, or 3200 hours, a yellow warning is triggered; below 50%, or 2000 hours, an orange warning is triggered; below 20%, or 800 hours, a red warning is triggered, and the pipeline shutoff valve is closed. The industrial monitoring system also notifies operators of the remaining life, warning level, and replacement recommendations via audible and visual alarms and text messages.

[0049] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A method for predicting the life of a bursting disc based on intelligent perception, characterized in that: The following steps are involved: S1. Construct a multimodal smart sensor array comprising an acoustic emission sensor, a strain sensor, a temperature sensor, and a pressure sensor. The multimodal smart sensor array is installed on the bursting disc and key surrounding locations according to a preset layout to collect real-time acoustic emission signals, strain information, temperature changes, and pressure fluctuation data during the operation of the bursting disc; S2. Establish a data transmission network to transmit data collected by the multimodal intelligent sensor array to the data processing center in real time through a wireless transmission module, while encrypting the data; S3. Preprocess the collected real-time data, using an adaptive filtering algorithm to remove noise based on the dynamic characteristics of the data. Standardization methods are used to unify data from different sensor types into a specific range to form a standardized data set. S4. Construct a deep learning model consisting of a cascade of a convolutional neural network and a long short-term memory network. Input the standardized dataset into the deep learning model. The convolutional neural network extracts spatial features from the data, while the long short-term memory network analyzes the temporal characteristics of the data, mining the data for potential features related to the lifespan of the bursting disc. S5. Based on the characteristic data output by the deep learning model and the fatigue damage accumulation theory, the formula Calculate the cumulative fatigue damage D of the bursting disc, where Δσ i is the stress amplitude of the i-th stress cycle, m is the material constant, C is a constant related to the material, and n is the number of stress cycles; S6. Based on the accumulated fatigue damage of the bursting disc and the preset life threshold, the remaining life of the bursting disc is predicted. When the remaining life is lower than the set safety limit, a life warning message is generated.

2. The method for predicting the life of a bursting disc based on intelligent perception according to claim 1, characterized in that: The preset layout is determined based on the structural characteristics and stress distribution of the bursting disc. The acoustic emission sensor is installed at the location of the bursting disc where cracks are likely to occur, the strain sensor is arranged in the stress concentration area, and the temperature sensor and pressure sensor are installed on the medium flow path close to the bursting disc.

3. The method for predicting the life of a bursting disc based on intelligent perception according to claim 1, characterized in that: In the step of establishing a data transmission network, the wireless transmission module adopts low-power wide area network technology to ensure stable data transmission in a complex industrial environment, and the data encryption processing adopts a combination of symmetric encryption and asymmetric encryption to improve the security of data transmission.

4. The method for predicting the life of a bursting disc based on intelligent perception according to claim 1, characterized in that: The adaptive filtering algorithm is based on the minimum mean square error criterion and tracks the noise changes in the data in real time by continuously adjusting the filter coefficients to achieve effective noise suppression.

5. The method for predicting the life of a bursting disc based on intelligent perception according to claim 1, characterized in that: After building the deep learning model, the historical failure data and normal operation data are used to train the deep learning model. During the training process, the stochastic gradient descent algorithm is combined with the momentum factor to adjust the model parameters to minimize the prediction error. The prediction error is calculated by the mean square error formula. Calculate, where N is the number of samples, y j is the actual fatigue damage accumulation, is the fatigue damage accumulation predicted by the model.

6. The method for predicting the life of a bursting disc based on intelligent perception according to claim 1, characterized in that: In the step of predicting the remaining life of the bursting disc, a mapping relationship model between the remaining life and the fatigue damage accumulation is established. By comparing the position of the current fatigue damage accumulation in the mapping relationship model, the remaining life of the bursting disc is determined. The mapping relationship model is an exponential decay model L 剩余 =L0·e -kD , where L0 is the initial life and k is the attenuation coefficient, which is determined by fitting historical data.

7. The method for predicting the life of a bursting disc based on intelligent perception according to claim 1, characterized in that: The life warning information includes the remaining life value, warning level and replacement suggestion, which is sent to relevant operators through the industrial monitoring system in the form of sound and light alarms and SMS notifications. The warning level includes three threshold settings: When the remaining life prediction value is lower than 80% of the design life, a yellow warning is triggered; When the remaining life prediction value is lower than 50% of the design life, an orange warning is triggered; When the remaining life prediction value is lower than 20% of the design life, a red warning is triggered and the equipment is shut down for protection.

8. The method for predicting the life of a bursting disc based on intelligent perception according to claim 1, characterized in that: Regularly calibrate and maintain the multimodal intelligent sensor array, and correct the collected data based on the calibration results to ensure the accuracy and reliability of the data. At the same time, update the training data of the deep learning model and optimize the model prediction performance.

Images