A method for predicting the remaining useful life of industrial bearings based on big data

Abstract

The present invention belongs to the field of deep learning, and specifically relates to a method for predicting the remaining service life of industrial bearings based on big data, including: collecting signals during the operation of bearings from multiple rolling bearings, where the signals include vibration signals, temperature, and load; performing denoising and normalization processing on the collected signals; inputting the standardized signals into a time series prediction network to extract the deep features of the signals; inputting the deep features into a hierarchical cross-convolution network to obtain a forward information representation and a reverse information representation; using an information fusion technology to fuse the forward information representation and the reverse information representation, and using a head and tail needle technology to process the fused information representation to obtain a corresponding position information representation; outputting a prediction result of the remaining service life of the industrial bearing according to the corresponding position information representation; the present invention solves the problem of predicting the remaining service life of industrial bearings.

Description

Technical Field

[0001] The present invention belongs to the field of deep learning, and particularly relates to a method for predicting the remaining service life of industrial bearings based on big data. Background Art

[0002] With the rapid development of industrial automation and intelligence, the health management and fault prediction of mechanical equipment have increasingly become the focus of attention in the industrial and academic fields. In industrial production, various mechanical equipment plays a crucial role, and their smooth operation directly affects the stability and efficiency of the production line. Among them, rolling bearings, as a common rotating component, undertake the important function of supporting and transmitting the load of rotating components. However, due to factors such as harsh working environments, load changes, and long-term wear, rolling bearings are prone to damage and faults, which in turn affect the operating efficiency and safety of the entire mechanical system.

[0003] Accurately predicting the remaining useful life (RUL) of rolling bearings is crucial for taking timely maintenance measures, reducing production line downtime, and improving production efficiency. Traditional remaining life prediction models based on empirical rules or simple statistical methods often fail to meet the requirements of accuracy and robustness in actual production. Therefore, more accurate and reliable methods for predicting the remaining service life of industrial bearings are needed to address the challenges. Summary of the Invention

[0004] To solve the above problems existing in the prior art, the present invention proposes a method for predicting the remaining service life of industrial bearings based on big data. The method includes: obtaining signals generated by the industrial bearings to be detected, inputting the signals into a trained prediction model for the remaining service life of industrial bearings, and obtaining a prediction result;

[0005] Training the prediction model for the remaining service life of industrial bearings includes: collecting signals during the operation of bearings from multiple rolling bearings, where the signals include vibration signals, temperature, and load; performing denoising and normalization processing on the collected signals; inputting the standardized signals into a time series prediction network to extract the deep features of the signals; inputting the deep features into a hierarchical cross-convolution network to obtain a forward information representation and a reverse information representation; using an information fusion technology to fuse the forward information representation and the reverse information representation, and using a head and tail needle technology to process the fused information representation to obtain a corresponding position information representation; outputting a prediction result for the remaining service life of industrial bearings according to the corresponding position information representation; calculating the loss function of the model according to the prediction result, adjusting the model parameters, and completing the training of the model when the loss function converges.

[0006] Advantages of the Present Invention

[0007] The present invention proposes an innovative method based on big data and deep learning for predicting the remaining service life of industrial bearings. The core lies in using various operating data such as vibration signals, temperature, and load of multiple bearings, and through a series of complex data processing and model training steps, achieving accurate prediction of the remaining life of the bearings. The present invention not only solves the problem of predicting the remaining service life of industrial bearings, but also provides a reliable and efficient prediction method for the industrial field, and is expected to play an important role in industrial production, improving production efficiency and equipment operation stability. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] Figure 1 is a flowchart of a method for predicting the remaining service life of industrial bearings based on big data according to the present invention;

[0009] Figure 2 is a flowchart of model training for a method for predicting the remaining service life of industrial bearings based on big data according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0010] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0011] A method for predicting the remaining service life of industrial bearings based on big data includes: acquiring signals generated by the industrial bearings to be detected, and inputting the signals into a trained prediction model for the remaining service life of industrial bearings to obtain a prediction result.

[0012] A method for predicting the remaining service life of industrial bearings based on big data, as Figure 1 shown, this method includes: collecting operating data from rolling bearings, and sending the preprocessed data into a time series test network (STN); sending the deep features into a hierarchical cross convolutional network (LCCN) to output the representations of forward and reverse information; combining the representations of forward and reverse information to obtain the representation of relative position information and output the model prediction result; calculating the loss of the model prediction result and training by repeated iteration.

[0013] As Figure 2 shown, training the prediction model for the remaining service life of industrial bearings includes:

[0014] S1. Collect operating data such as vibration signals, temperature, and load from multiple rolling bearings, and perform denoising and standardization processing.

[0015] The steps for data denoising and standardization include:

[0016] S11. Check the sensor data, process outliers, duplicate values, and missing values to ensure data integrity.

[0017] S12. Perform complex waveform transformation on the sensor data to remove noise in the data and retain important signal features.

[0018] S13. Standardize the denoised data to eliminate the dimensional and amplitude differences between different sensors.

[0019] In this embodiment, the calculation rule for removing noise from sensor data is:

[0020]

[0021] where x denoised (t) is the denoised signal; J is the maximum scale of the complex waveform transformation, ω j,k is the waveform transformation coefficient, representing the components of the signal at different scales and positions, is the waveform transformation function, describing the waveform transformation shape at different scales and positions.

[0022] The calculation rules for standardizing the denoised data include:

[0023]

[0024] where x standardized (t) is the standardized signal; x denoised (t) is the denoised signal; μ is the mean of the denoised signal, σ is the standard deviation of the denoised signal, α is the parameter controlling the hyperbola shape, and β is the offset parameter.

[0025] S2. Feed the preprocessed data into the Sequential Temporal Network (STN), and use the attention mechanism to extract the deep features of the data. Specifically, it includes:

[0026] S21. Convert the processed data into time series data.

[0027] S22. Through the attention mechanism, perform weighted processing on the serialized data. The calculation rules for performing weighted processing on the data include:

[0028]

[0029] where 1 / Z is the normalization factor; σ is an activation function, W is the weight matrix; x i is the i-th data point, and b is the bias term.

[0030] S23. Feed the weighted data into the Sequential Temporal Network (STN) to extract the deep features of the data.

[0031] S3. Feed the depth features into the Layered Cross Convolutional Network (LCCN) to obtain the structural dependencies between components and output the representations of forward and reverse information. Specifically, it includes:

[0032] S31. Use multiple convolutional layers to directly transfer the low-level features to the high-level.

[0033] S32. Through the skip connection mechanism, enable the features of different levels to exchange and fuse with each other. The calculation rules of the skip connection mechanism include:

[0034] Y (l) = ReLU(X (l) + X (l-k) )

[0035] In the formula, ReLU is the rectified linear unit activation function, Y(l) is the feature representation after the skip connection in the l-th layer, X (l) represents the feature input to the l-th layer, l represents the number of layers of the network, and k is the step size of the skip.

[0036] S33. Use the concatenation and weighted average method to obtain the comprehensive feature representation. The calculation rules for concatenation and weighting to obtain the comprehensive feature representation include:

[0037] F final = α·F original +(1 - α)·F skipconnected

[0038] In the formula, F original is the original feature map, F skipconnected is the processed feature map, and α is the weight of weighted concatenation.

[0039] S34. Input the comprehensive feature representation into the fully connected layer for learning and output the representations of forward and reverse information.

[0040] S4. Combine the representations of forward and reverse information through information fusion technology, adopt the head and tail pointer technology to obtain the representation of relative position information, and output the model prediction result. Specifically, it includes:

[0041] S41. Scan the representations of forward and reverse information with a sliding window and encode the relative position vector using the cosine function.

[0042] S42. Use the pooling technology to extract information from the head and tail of the relative position. The calculation rules of the pooling technology include:

[0043]

[0044] In the formula, x iis the encoded relative position vector, N is the length of the relative position vector, α is used to control the attention to the head and tail, and σ maps x i to the interval (0, 1).

[0045] S43. Perform dynamic time alignment on the extracted head and tail information to extract the representation of the relative position information.

[0046] The calculation rules adopted by the information entropy loss calculator include:

[0047]

[0048] where N is the number of categories of the prediction results, p i is the prediction probability of the i-th category, α i is a regulation parameter related to the category, β i is another regulation parameter related to the category, and γ i is the normalization factor of the i-th category.

[0049] The above-mentioned embodiments further elaborate on the purpose, technical solutions, and advantages of the present invention. It should be understood that the above-mentioned embodiments are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made to the present invention within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for predicting the remaining service life of industrial bearings based on big data, characterized in that: include: Acquire the signal generated by the industrial bearing to be tested, input the signal into the trained industrial bearing remaining service life prediction model, and obtain the prediction result; Training the remaining service life prediction model for industrial bearings includes: collecting signals from multiple rolling bearings when the bearings are working, including vibration signals, temperature and load; denoising and standardizing the collected signals; inputting the standardized signals into the timing prediction network to extract the deep features of the signals; inputting the deep features into the layered cross convolution network to obtain forward information representation and reverse information representation; using information fusion technology to fuse the forward information representation and the reverse information representation, and using the head-tail needle technology to process the fused information representation to obtain the corresponding position information representation; outputting the prediction results of the remaining service life of the industrial bearings according to the corresponding position information representation; calculating the loss function of the model according to the prediction results, adjusting the model parameters, and completing the model training when the loss function converges; The fusion of forward information representation and reverse information representation includes: using a sliding window to scan the forward and reverse information representation, using a cosine function to encode the scanned forward and reverse information representation to obtain a relative position vector; using pooling technology to extract information from the head and tail of the relative position vector; dynamically aligning the extracted head and tail information to extract the representation of the relative position information.

2. The method for predicting the remaining service life of industrial bearings based on big data according to claim 1 is characterized in that: The denoising and standardization processing of the collected signals includes: de-duplication processing of the collected signals, that is, deleting abnormal values, duplicate values ​​and missing values ​​in the signals; performing complex waveform transformation on the de-duplication signal to remove noise in the signal; and standardizing the denoised signal.

3. The method for predicting the remaining service life of industrial bearings based on big data according to claim 2 is characterized in that: The complex waveform transformation of the signal after redundancy removal includes: Among them, x denoised (t) is the denoised signal; J is the maximum scale of the complex transformation of the waveform, ω j,k is the waveform transformation coefficient, is the waveform transformation function.

4. The method for predicting the remaining service life of industrial bearings based on big data according to claim 2 is characterized in that: The denoised signal is standardized as follows: Among them, x standardized (t) is the normalized signal; x denoised (t) is the denoised signal; μ is the mean of the denoised signal, σ is the standard deviation of the denoised signal, α is the parameter controlling the shape of the hyperbola, and β is the offset parameter.

5. The method for predicting the remaining service life of industrial bearings based on big data according to claim 1 is characterized in that: The time series prediction network processes the input signal by: converting the standardized signal into time series data; using the attention mechanism to perform weighted processing on the time series data; and inputting the weighted data into the time series prediction network to obtain the deep features of the data.

6. The method for predicting the remaining service life of industrial bearings based on big data according to claim 5 is characterized in that: Weighting of time series data includes: Among them, 1 / Z is the normalization factor; σ is an activation function, W is the weight matrix; x i is the ith data point and b is the bias term.

7. The method for predicting the remaining service life of industrial bearings based on big data according to claim 1 is characterized in that: The layered cross convolutional network processes deep features by: using multiple convolutional layers to directly transfer low-level features to high-level features; using a skip connection mechanism to exchange and fuse features at different levels; concatenating and weighted averaging the exchanged and fused features to obtain a comprehensive feature representation; and inputting the comprehensive feature representation into the fully connected layer for learning to obtain forward information representation and reverse information representation.

8. The method for predicting the remaining service life of industrial bearings based on big data according to claim 1 is characterized in that: The pooling technique is used to extract information from the head and tail of the relative position vector: Among them, x i is the relative position vector of the code, N is the length of the relative position vector, α is the degree of attention to the head and tail, and σ is x i Mapped to the interval (0,1).

9. The method for predicting the remaining service life of industrial bearings based on big data according to claim 1 is characterized in that: The expression of the loss function is: Among them, N is the number of categories of prediction results, p i is the predicted probability of the i-th category, α i is a class-dependent tuning parameter, β i is another class-dependent tuning parameter, γ i is the normalization factor for the ith category.

Images