Intelligent diagnosis method for stator base bolt based on optical fiber strain

Abstract

The invention discloses a stator base bolt intelligent diagnosis method based on optical fiber strain, and relates to the technical field of equipment state monitoring, and the method comprises the steps: collecting original strain data through an optical fiber sensor disposed on a stator base bolt, carrying out the preprocessing and segmentation processing of the original strain data, and obtaining the preprocessed strain data; performing multi-scale feature extraction based on the preprocessed strain data, and constructing a strain feature vector; and performing dynamic weight adjustment on the strain feature vector according to real-time environment temperature and load data, and performing bolt health state prediction based on the adjusted strain feature vector. According to the method, the description of the stress state of the bolt can be more comprehensive, the influence of environmental disturbance on a diagnosis result is reduced, the judgment of the bolt state is more accurate and reliable, and the accuracy, robustness and adaptability of health assessment of the stator frame bolt are improved.

Description

Technical Field

[0001] This invention relates to the field of equipment condition monitoring technology, and in particular to an intelligent diagnostic method for stator frame bolts based on fiber optic strain. Background Technology

[0002] With the increasing demands for structural stability in large hydropower units and other heavy-duty rotating machinery, the health status monitoring technology for stator frame bolts is gradually evolving from traditional manual inspections and periodic tightening maintenance towards real-time, online, and intelligent methods. In recent years, with the rapid development of fiber optic sensing technology, strain monitoring technology, and signal analysis methods, structural state sensing methods based on fiber optic gratings (FBGs) and fiber optic distributed sensing have been widely applied to stress and deformation measurement in electromechanical equipment. Fiber optic sensors possess advantages such as resistance to electromagnetic interference, high sensitivity, strong environmental resistance, and fast response speed, making them particularly suitable for deployment in structural parts with complex stress and limited space, such as motor stator frame bolts. In terms of signal processing, multi-scale feature extraction, frequency domain analysis, and wavelet transform methods are gradually becoming core technologies for strain diagnosis, enabling more accurate capture and quantification of minute changes in equipment operating status. Simultaneously, the development of online learning algorithms, such as incremental support vector machines and recurrent neural networks, allows equipment health diagnosis to continuously update models in dynamic operating environments, improving the adaptability and stability of prediction results.

[0003] However, existing stator frame bolt condition monitoring technologies still face several shortcomings. On the one hand, traditional strain monitoring relies heavily on single-scale features or a single sensing mode, failing to effectively reflect the true stress state of bolts under complex loads and temperature disturbances, resulting in insufficient diagnostic accuracy. On the other hand, most diagnostic methods lack comprehensive consideration of external factors such as real-time ambient temperature and unit load fluctuations, making strain characteristics highly sensitive to environmental changes and prone to misjudgment. Furthermore, existing online learning models often cannot dynamically adjust weights based on real-time error information, making it difficult for the models to adapt to long-term disturbances such as equipment aging and changes in operating conditions, leading to a decline in predictive performance over time. Summary of the Invention

[0004] In view of the problems existing in the current intelligent diagnostic method for stator frame bolts based on fiber optic strain, this invention is proposed. Therefore, the problem to be solved by this invention is how to provide an intelligent diagnostic method for stator frame bolts based on fiber optic strain.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0006] In a first aspect, the present invention provides an intelligent diagnostic method for stator frame bolts based on fiber optic strain, comprising: acquiring raw strain data by means of fiber optic sensors arranged on stator frame bolts, and preprocessing and segmenting the raw strain data to obtain preprocessed strain data.

[0007] Multi-scale feature extraction is performed on the preprocessed strain data to construct a strain feature vector;

[0008] The strain feature vector is dynamically weighted based on real-time ambient temperature and load data, and the bolt health status is predicted based on the adjusted strain feature vector.

[0009] As a preferred embodiment of the intelligent diagnostic method for stator frame bolts based on fiber optic strain described in this invention, the method for obtaining preprocessed strain data includes:

[0010] Calculate the baseline value for the data of each sensor under static initial conditions and subtract the baseline value from subsequent sampling;

[0011] A digital bandpass filter is applied to the calibrated raw signal to suppress noise in the external frequency band, and the filtering result is smoothed.

[0012] The filtered and smoothed data from each sensor are normalized, and the minimum-maximum scaling method is used to convert the data to a uniform dimension.

[0013] The preprocessed data is segmented according to time windows, and the statistical characteristics of each time period are extracted to obtain the preprocessed strain data.

[0014] As a preferred embodiment of the intelligent diagnostic method for stator frame bolts based on fiber optic strain described in this invention, the construction of the strain feature vector includes:

[0015] In the time domain, the preprocessed strain data are statistically described, and the arithmetic mean, variance, root mean square, kurtosis, skewness, and peak-to-peak value are calculated to obtain the time domain characteristics.

[0016] The preprocessed strain data is subjected to fast Fourier transform to obtain the spectral amplitude sequence. The main frequency component is determined according to the position of the amplitude peak, and the spectral energy, spectral centroid and bandwidth characteristics are calculated to obtain the frequency domain characteristics.

[0017] Wavelet decomposition is performed on the preprocessed strain data to decompose the original sequence into several levels of approximation coefficients and detail coefficients. The root mean square, energy, variance and mean of each level of coefficients are calculated to obtain wavelet domain features.

[0018] The time-domain features, frequency-domain features, and wavelet-domain features are concatenated into an initial vector in a fixed order to obtain the time-domain feature vector, frequency-domain feature vector, and wavelet-domain feature vector, respectively.

[0019] The feature vectors from different domains are combined into a multi-scale strain feature vector through a weighted linear fusion method. The fusion expression is as follows:

[0020] ;

[0021] in: This is a multi-scale strain eigenvector. For time-domain feature vectors, For frequency domain eigenvectors, For wavelet domain eigenvectors, , and This is the fusion coefficient.

[0022] As a preferred embodiment of the intelligent diagnostic method for stator frame bolts based on fiber optic strain described in this invention, the step of dynamically adjusting the weights of the strain feature vector according to real-time ambient temperature and load data includes:

[0023] Synchronize and align the real-time ambient temperature and real-time external load with the obtained strain eigenvector.

[0024] Based on the sensitivity coefficient set of each feature to temperature and load obtained from offline calibration, the comprehensive influence factor of the feature is calculated and expressed as:

[0025] ;

[0026] in: Indicates the temperature deviation on the first The linear sensitivity of each feature Indicates the load deviation on the first The linear sensitivity of each feature For the first A comprehensive influencing factor of each characteristic;

[0027] The benchmark weights are dynamically adjusted, as follows:

[0028] ;

[0029] in: For the adjusted number Each feature weight, For the first The baseline weights of each feature This is the scaling factor;

[0030] The weighted feature vector is generated using the adjusted feature weights, and is represented as follows:

[0031] in: These are weighted eigenvectors.

[0032] As a preferred embodiment of the intelligent diagnostic method for stator frame bolts based on fiber optic strain described in this invention, the bolt health status prediction based on the adjusted strain feature vector includes:

[0033] The weighted feature vectors are input into the online incremental learning model to predict health status. The model pre-training step is completed before deployment. An initial classifier is trained using a labeled historical sample set, with categories including normal tightness, slight looseness, and severe damage.

[0034] The online model uses incremental support vector machines to predict the health status of bolts in multiple categories, mapping the three health statuses to three independent binary classifiers.

[0035] For each health category, an incremental support vector machine classifier is constructed. For the input weighted feature vector, each binary classifier provides a corresponding decision function value, expressed as:

[0036] ;

[0037] in: For the first The decision function value of the category, For the first SVM weight vectors for each category For bias terms, This is the transpose symbol.

[0038] As a preferred embodiment of the intelligent diagnostic method for stator frame bolts based on fiber optic strain described in this invention, the health status prediction further includes:

[0039] Mapping the decision function values ​​to class probabilities and then normalizing them yields the probabilities for each class, expressed as:

[0040] ;

[0041] ;

[0042] in: For the first Unnormalized probability of the class These are temperature control parameters, set during the model calibration phase. For the first The probability of the category, For the first The decision function value for the category;

[0043] The category with the highest probability is selected as the predicted category, and the stator frame bolt diagnosis is performed based on the obtained category probability.

[0044] As a preferred embodiment of the intelligent diagnostic method for stator frame bolts based on fiber optic strain described in this invention, the stator frame bolt diagnosis includes:

[0045] If the predicted category is Category 1 and the probability of Category 1 is greater than the first category threshold, the stator base bolt is diagnosed as severely damaged; if the predicted category is Category 2 and the probability of Category 2 is greater than the second category threshold, the stator base bolt is diagnosed as slightly loose; if the predicted category is Category 3 and the probability of Category 3 is greater than the third category threshold, the stator base bolt is diagnosed as normally tightened. The first, second, and third category thresholds are predetermined during the calibration phase.

[0046] In a second aspect, the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of a method for intelligent diagnosis of stator frame bolts based on fiber optic strain.

[0047] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements the steps of a method for intelligent diagnosis of stator frame bolts based on fiber optic strain.

[0048] The beneficial effects of this invention are as follows: This invention directly acquires strain changes caused by bolt stress through fiber optic sensors, achieving highly sensitive and interference-resistant in-situ monitoring, thus improving data stability and representativeness; multi-scale feature extraction integrates time-domain, frequency-domain, and structural change information, making the description of bolt stress state more comprehensive; dynamic adjustment of feature weights based on real-time temperature and load reduces the impact of environmental disturbances on diagnostic results; and health prediction based on the adjusted feature vectors makes bolt condition judgment more accurate and reliable. This improves the accuracy, robustness, and adaptability of stator base bolt health assessment. Attached Figure Description

[0049] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0050] Figure 1This is a flowchart of a smart diagnostic method for stator frame bolts based on fiber optic strain. Detailed Implementation

[0051] To make the above-mentioned objects, features, and advantages of the present invention more readily understood, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0052] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0053] Secondly, the term "one embodiment" or "example" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the invention. An embodiment appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single embodiment or an embodiment selectively mutually exclusive with other embodiments.

[0054] Reference Figure 1 This is the first embodiment of the present invention, which provides a smart diagnostic method for stator frame bolts based on fiber optic strain, comprising:

[0055] S1: Raw strain data is collected by fiber optic sensors arranged on the stator frame bolts, and the raw strain data is preprocessed and segmented to obtain preprocessed strain data.

[0056] S2: Based on the preprocessed strain data, perform multi-scale feature extraction to construct a strain feature vector;

[0057] S3: Dynamically adjust the weights of the strain feature vector based on real-time ambient temperature and load data, and predict the health status of the bolts based on the adjusted strain feature vector.

[0058] Specifically, multiple fiber optic strain sensors are embedded in the surface of the stator frame bolts or in the bolt holes and synchronized with the data acquisition unit to acquire the original strain signal sequence.

[0059] The acquired raw strain signals are first baseline calibrated: the baseline value is calculated for several seconds of data from each sensor under static initial conditions and the baseline is subtracted from subsequent sampling to eliminate installation offset and static strain bias.

[0060] A digital bandpass filter is applied to the calibrated raw signal to suppress noise in the external frequency band. Finally, the filtering result is smoothed, and the filtered and smoothed data of each sensor is normalized. Minimum-maximum scaling is used for transformation to obtain the preprocessed strain data.

[0061] Multi-scale features are extracted sequentially from the preprocessed strain data. Statistical descriptions are performed on the preprocessed strain data in the time domain, calculating the arithmetic mean, variance, root mean square (RMS), kurtosis, skewness, and peak-to-peak value. Subsequently, a Fast Fourier Transform (FFT) is performed on the preprocessed strain data to obtain a spectral amplitude sequence. The dominant frequency component is determined based on the position of the amplitude peak, and features such as spectral energy, spectral centroid, and bandwidth are calculated. Wavelet decomposition is then performed on the preprocessed strain data, decomposing the original sequence into several levels of approximation coefficients and detail coefficients. The RMS, energy, variance, and mean of each level of coefficients are calculated.

[0062] The time-domain features, frequency-domain features, and wavelet-domain features are concatenated into an initial vector in a fixed order.

[0063] Multi-scale feature fusion is performed on the concatenated initial vector. Features from different domains are combined into a multi-scale strain feature vector through weighted linear fusion. The fusion expression is as follows:

[0064] ;

[0065] in: This is a multi-scale strain eigenvector. For time-domain feature vectors, For frequency domain eigenvectors, For wavelet domain eigenvectors, , and This is the fusion coefficient, used to adjust the importance of features from different domains.

[0066] The fused vector is the multi-scale strain feature vector for each strain data segment.

[0067] During runtime, the system synchronizes and aligns the real-time ambient temperature and real-time external load with the obtained strain feature vector, calculates the comprehensive impact factor, dynamically adjusts the feature weights, and provides online health predictions.

[0068] Preset calibration baseline temperature, calibration baseline load, temperature calibration range, and load calibration range;

[0069] Based on the sensitivity coefficient set of each feature to temperature and load obtained from offline calibration (determined through historical calibration data or experiments), the comprehensive influence factor of the feature is calculated and expressed as:

[0070] ;

[0071] in: Indicates the temperature deviation on the first The linear sensitivity of each feature Indicates the load deviation on the first The linear sensitivity of each feature For the first The comprehensive influencing factor of each characteristic.

[0072] A predefined set of baseline weights for each feature (the baseline weights are derived from previous feature importance analysis or expert settings) is used to dynamically adjust the baseline weights, as shown below:

[0073] ;

[0074] in: For the adjusted number Each feature weight, For the first The baseline weights of each feature Scaling factor >0 and determined during the system calibration phase.

[0075] The weighted feature vector is generated using the adjusted feature weights, and is represented as follows:

[0076] ;

[0077] in: These are weighted eigenvectors;

[0078] The weighted feature vectors are input into the online incremental learning model for health status prediction. The model pre-training step is completed before deployment: an initial classifier is trained using a labeled historical sample set, with categories including normal tightness, slight looseness, and severe damage. The model parameters are then loaded. The online model supports two implementation methods and operates through a unified interface.

[0079] Incremental Support Vector Machine (iSVM) implementation: Maintains a set of support vectors with a finite budget using an incremental update algorithm with kernelized or linear kernels.

[0080] The confidence score is obtained by performing a decision function on the input weighted feature vector (either through a one-to-many SVM or by converting it into class probability through probabilistic processing).

[0081] The online model uses incremental support vector machines to predict the health status of bolts in multiple categories. It maps the three health statuses to three independent binary classifiers, each of which distinguishes between its own class and other classes.

[0082] A set of predefined health categories is used, corresponding to normal tightness, slight looseness, and severe damage. For each health category, an incremental support vector machine classifier is constructed and trained to distinguish the sample category.

[0083] For the input weighted feature vector, each binary classifier provides a corresponding decision function value, expressed as:

[0084] ;

[0085] in: For the first The decision function value of the category, For the first SVM weight vectors for each category For bias terms, It is the transpose symbol;

[0086] The decision function values ​​are mapped to class probabilities and normalized to obtain the probabilities of each class. The class with the highest probability is selected as the predicted class, expressed as:

[0087] ;

[0088] ;

[0089] in: For the first Unnormalized probability of the class These are temperature control parameters, set during the model calibration phase. For the first The probability of the category, For the first The decision function value for the category;

[0090] The stator frame bolts are diagnosed based on the obtained category probabilities. If the predicted category is Category 1 and the probability of Category 1 is greater than the first category threshold, the stator frame bolts are diagnosed as severely damaged. If the predicted category is Category 2 and the probability of Category 2 is greater than the second category threshold, the stator frame bolts are diagnosed as slightly loose. If the predicted category is Category 3 and the probability of Category 3 is greater than the third category threshold, the stator frame bolts are diagnosed as normally tightened.

[0091] Among them, the thresholds for the first category, the second category, and the third category are all predetermined during the calibration phase.

[0092] On the monitoring interface, the online prediction results are projected in layers according to risk level: the prediction category is mapped to a three-layer display (normal tightness → green layer; slight looseness → yellow layer; severe damage → red layer).

[0093] This embodiment also provides a computer device applicable to a method for intelligent diagnosis of stator frame bolts based on fiber optic strain, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement all or part of the steps of the method described in the above embodiments of the present invention.

[0094] This embodiment also provides a storage medium storing a computer program thereon. When the computer program is executed by a processor, it performs the method in any optional implementation of the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read-Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0095] The storage medium proposed in this embodiment and the data storage method proposed in the above embodiments belong to the same inventive concept. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0096] In summary, this invention directly acquires strain changes caused by bolt stress using fiber optic sensors, achieving highly sensitive and interference-resistant in-situ monitoring. Preprocessing and segmented processing improve data stability and representativeness. Multi-scale feature extraction integrates time-domain, frequency-domain, and structural change information, providing a more comprehensive description of bolt stress state. Dynamic adjustment of feature weights based on real-time temperature and load significantly reduces the impact of environmental disturbances on diagnostic results. Health prediction based on the adjusted feature vectors makes bolt condition assessment more accurate and reliable. This improves the accuracy, robustness, and adaptability of stator base bolt health assessment.

[0097] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A smart diagnostic method for stator frame bolts based on fiber optic strain, characterized in that: include, Raw strain data is collected by fiber optic sensors arranged on the stator frame bolts, and the raw strain data is preprocessed and segmented to obtain preprocessed strain data. Multi-scale feature extraction is performed on the preprocessed strain data to construct a strain feature vector; The strain feature vector is dynamically weighted based on real-time ambient temperature and load data, and the bolt health status is predicted based on the adjusted strain feature vector.

2. The intelligent diagnostic method for stator frame bolts based on fiber optic strain as described in claim 1, characterized in that: The obtained preprocessed strain data includes: Calculate the baseline value for the data of each sensor under static initial conditions and subtract the baseline value from subsequent sampling; A digital bandpass filter is applied to the calibrated raw signal to suppress noise in the external frequency band, and the filtering result is smoothed. The filtered and smoothed data from each sensor are normalized, and the minimum-maximum scaling method is used to convert the data to a uniform dimension. The preprocessed data is segmented according to time windows, and the statistical characteristics of each time period are extracted to obtain the preprocessed strain data.

3. The intelligent diagnostic method for stator frame bolts based on fiber optic strain as described in claim 1, characterized in that: The construction of the strain feature vector includes: In the time domain, the preprocessed strain data are statistically described, and the arithmetic mean, variance, root mean square, kurtosis, skewness, and peak-to-peak value are calculated to obtain the time domain characteristics. The preprocessed strain data is subjected to fast Fourier transform to obtain the spectral amplitude sequence. The main frequency component is determined according to the position of the amplitude peak, and the spectral energy, spectral centroid and bandwidth characteristics are calculated to obtain the frequency domain characteristics. Wavelet decomposition is performed on the preprocessed strain data to decompose the original sequence into several levels of approximation coefficients and detail coefficients. The root mean square, energy, variance and mean of each level of coefficients are calculated to obtain wavelet domain features. The time-domain features, frequency-domain features, and wavelet-domain features are concatenated into an initial vector in a fixed order to obtain the time-domain feature vector, frequency-domain feature vector, and wavelet-domain feature vector, respectively. The feature vectors from different domains are combined into a multi-scale strain feature vector through a weighted linear fusion method. The fusion expression is as follows: ； in: This is a multi-scale strain eigenvector. For time-domain feature vectors, For frequency domain eigenvectors, For wavelet domain eigenvectors, , and This is the fusion coefficient.

4. The intelligent diagnostic method for stator frame bolts based on fiber optic strain as described in claim 1, characterized in that: The dynamic weight adjustment of the strain feature vector based on real-time ambient temperature and load data includes: Synchronize and align the real-time ambient temperature and real-time external load with the obtained strain eigenvector. Based on the sensitivity coefficient set of each feature to temperature and load obtained from offline calibration, the comprehensive influence factor of the feature is calculated and expressed as: ； in: Indicates the temperature deviation on the first The linear sensitivity of each feature Indicates the load deviation on the first The linear sensitivity of each feature For the first A comprehensive influencing factor of each characteristic; The benchmark weights are dynamically adjusted, as follows: ； in: For the adjusted number Each feature weight, For the first The baseline weights of each feature This is the scaling factor; The weighted feature vector is generated using the adjusted feature weights, and is represented as follows: ； in: These are weighted eigenvectors.

5. The intelligent diagnostic method for stator frame bolts based on fiber optic strain as described in claim 4, characterized in that: The bolt health status prediction based on the adjusted strain feature vector includes: The weighted feature vectors are input into the online incremental learning model to predict health status. The model pre-training step is completed before deployment. An initial classifier is trained using a labeled historical sample set, with categories including normal tightness, slight looseness, and severe damage. The online model uses incremental support vector machines to predict the health status of bolts in multiple categories, mapping the three health statuses to three independent binary classifiers. For each health category, an incremental support vector machine classifier is constructed. For the input weighted feature vector, each binary classifier provides a corresponding decision function value, expressed as: ； in: For the first The decision function value of the category, For the first SVM weight vectors for each category For bias terms, This is the transpose symbol.

6. The intelligent diagnostic method for stator frame bolts based on fiber optic strain as described in claim 1, characterized in that: The health status prediction also includes: Mapping the decision function values ​​to class probabilities and then normalizing them yields the probabilities for each class, expressed as: ； ； in: For the first Unnormalized probability of the class These are temperature control parameters, set during the model calibration phase. For the first The probability of the category, For the first The decision function value for the category; The category with the highest probability is selected as the predicted category, and the stator frame bolt diagnosis is performed based on the obtained category probability.

7. The intelligent diagnostic method for stator frame bolts based on fiber optic strain as described in claim 6, characterized in that: The stator frame bolt diagnosis includes: If the predicted category is Category 1 and the probability of Category 1 is greater than the first category threshold, the stator base bolt is diagnosed as severely damaged; if the predicted category is Category 2 and the probability of Category 2 is greater than the second category threshold, the stator base bolt is diagnosed as slightly loose; if the predicted category is Category 3 and the probability of Category 3 is greater than the third category threshold, the stator base bolt is diagnosed as normally tightened. The first, second, and third category thresholds are predetermined during the calibration phase.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the intelligent diagnostic method for stator frame bolts based on fiber optic strain as described in any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the intelligent diagnostic method for stator frame bolts based on fiber optic strain as described in any one of claims 1 to 7.

Images