A gas turbine sensor fault diagnosis method and system based on visual enhancement
Through Gram angle field transformation and directional gradient histogram extraction feature descriptor, combined with multi-classification support vector machine, the problem of limited signal feature extraction in gas turbine sensor fault diagnosis is solved, and the accuracy and reliability of fault signal classification is improved. Effective identification of micro fault signals is achieved.
Patent Information
- Application Number
- CN202211074831.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-09-02
- Publication Date
- 2024-11-22
- Estimated Expiration
- 2042-09-02
AI Technical Summary
In the fault diagnosis of gas turbine sensors, signal feature extraction is limited, making it difficult to effectively identify small amplitude fault signals. Especially under variable operating conditions, the fault classification effect is poor.
Gram angle field transformation and directional gradient histogram extraction feature descriptors are used to establish a fault diagnosis model through a multi-classification support vector machine, and signal feature extraction and classification classification are improved to improve the characterization ability and classification accuracy of fault signals.
It improves the accuracy and reliability of the classification results of gas turbine multi-sensor fault signals, enhances the diagnosis ability of micro fault signals, and meets the real-time fault signal classification needs of industrial sites.
Smart Images

Figure CN115617010B_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of gas turbine control system fault detection, and in particular to a gas turbine sensor fault diagnosis method based on visual enhancement. Background Art
[0002] As global energy consumption rises, energy and environmental issues are becoming increasingly serious. The intelligentization of energy systems can effectively improve the regulation level of energy systems and improve the efficiency of energy systems. It is an effective measure to achieve my country's dual carbon and dual control goals. The intelligentization of energy systems requires the installation of a large number of sensors to detect key parameters of energy systems online in real time, so as to perceive the status of key components in the energy system in real time and realize real-time regulation, health assessment and fault diagnosis of energy systems.
[0003] However, sensors often work in harsh environments such as high temperature, high pressure, high humidity, and strong magnetic fields, which inevitably cause abnormal signals and aging in sensors. Taking three 9F gas turbines in a power plant in Shenzhen, my country as an example, through years of fault information statistics, it is found that sensor failures account for the highest proportion of all faults during the use of gas turbines. Gas turbine status monitoring is usually divided into four steps: abnormality detection, fault classification, fault isolation, and fault mitigation. Fault detection is a key step, and fault detection also helps to achieve intelligent control. It can be seen that conducting research on sensor fault detection and classification of gas turbine control systems is of great significance to ensure the safe operation of gas turbines.
[0004] The original signals collected by the sensor often contain noise, so direct fault classification is not effective, and the data features of the original signal are not very obvious. The commonly used data-based fault classification process is to extract signal features, and then use a classifier to determine whether a fault has occurred and the type of fault based on the classification results. Common methods for extracting signal features include signal decomposition, but for small-amplitude fault signals and variable operating conditions, the features obtained by signal decomposition are limited.
[0005] Therefore, for the fault diagnosis of gas turbine sensors, it is necessary to study the signal feature extraction, and then classify the fault signals according to the extracted features to improve the diagnostic ability of fault signals with small amplitudes. Summary of the invention
[0006] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and to provide a gas turbine sensor fault diagnosis method and system based on visual enhancement.
[0007] The purpose of the present invention can be achieved by the following technical solutions:
[0008] A gas turbine sensor fault diagnosis method based on visual enhancement comprises the following steps:
[0009] S1. Collecting sensor signals of the gas turbine to obtain an original signal time series, adding different sensor faults to the original signal time series to obtain a fault signal time series;
[0010] S2, convert the time series into a two-dimensional image using the Gram angular field transform, encode the time series values as angle cosines, and record the timestamp as the radius;
[0011] S3, using the oriented gradient histogram to extract feature descriptors in the two-dimensional image, taking the feature descriptors extracted from the original signal time series as normal signal samples, and taking the feature descriptors extracted from the fault signal time series as fault signal samples, to obtain a signal data set;
[0012] S4, establishing a multi-classification support vector machine as a fault diagnosis model, and using the signal data set to train the fault diagnosis model to obtain a trained fault diagnosis model;
[0013] S5. Obtain the sensor signal to be diagnosed, perform Gram angle field transformation and feature descriptor extraction, input the extracted feature descriptor into the trained fault diagnosis model, and obtain the fault diagnosis result.
[0014] Furthermore, step S2 is specifically as follows:
[0015] S21, obtain the time series to be converted X = {x1, x2, ..., x t ,…,x L}, where L represents the length of the time series, scaling the time series X to a uniform scale space;
[0016] S22. Encode the values of the time series as the cosine of the angle, with the timestamp as the radius, and re-express the scaled time series in polar coordinates using the following equation:
[0017]
[0018] In the formula, t i is the timestamp of the time series, N is a preset constant factor, It is a time series At time t i The value of
[0019] S23. Define the time correlation of different time intervals by calculating the trigonometric sum or trigonometric difference between each point, and obtain the Gram angular field matrix, that is, the two-dimensional image of the time series conversion.
[0020] Furthermore, in step S3, extracting feature descriptors in the two-dimensional image using the histogram of oriented gradients includes the following steps:
[0021] 1) Grayscale the two-dimensional image;
[0022] 2) Performing global image normalization on the two-dimensional image after grayscale processing;
[0023] 3) Calculate the gradient of each pixel, including size and direction information;
[0024] 4) Divide the entire two-dimensional image into multiple blocks, each block includes multiple cell units, and calculate the gradient histogram of each cell unit respectively;
[0025] 5) For each block, concatenate the gradient histograms of all cells in the block into a vector, normalize the vector, and obtain the feature descriptor of each block;
[0026] 6) Merge the feature descriptors of each block to obtain the feature descriptor of the entire image.
[0027] Furthermore, the signal data set is divided into a training set, a validation set and a test set, and the fault diagnosis model is trained based on the training set, the validation set and the test set to obtain a trained fault diagnosis model.
[0028] Furthermore, the training set, validation set and test set are unbalanced data sets, in which the ratio of normal signal samples to various types of fault signals is 10:3.
[0029] Furthermore, the loss function of the multi-classification support vector machine is as follows:
[0030]
[0031] In the formula, the first term is the regularization penalty, the second term is the multi-classification data loss, n represents the amount of data in the training set, and x i is the input data, y i is the label, W,b is the model parameter, φ(x i ) represents the kernel function, and C is the trade-off between the two items.
[0032] Furthermore, in step S1, adding different sensor faults to the original signal time series specifically includes: adding steps, short circuits, drifts, noises, and periodic signals of different amplitudes to the sensor signal.
[0033] A gas turbine sensor fault diagnosis system based on visual enhancement includes a data acquisition module, a Gram angle field transformation module, a feature descriptor extraction module and a classification module;
[0034] The data acquisition module is used to collect sensor signals to be diagnosed;
[0035] The Gram angular field transform module is used to convert the sensor time series to be diagnosed into a two-dimensional image using the Gram angular field transform, encode the value of the time series as the angle cosine, and record the time stamp as the radius;
[0036] The feature descriptor extraction module is used to extract feature descriptors in a two-dimensional image using a histogram of oriented gradients;
[0037] The classification module is equipped with a trained classification model, which is used to input the feature descriptor of the signal to be detected into the feature model to obtain the classification result as the fault diagnosis result;
[0038] The training process of the classification model includes the following steps:
[0039] S1. Collecting sensor signals of the gas turbine to obtain an original signal time series, adding different sensor faults to the original signal time series to obtain a fault signal time series;
[0040] S2, convert the time series into a two-dimensional image using the Gram angular field transform, encode the time series values as angle cosines, and record the timestamp as the radius;
[0041] S3, using the oriented gradient histogram to extract feature descriptors in the two-dimensional image, taking the feature descriptors extracted from the original signal time series as normal signal samples, and taking the feature descriptors extracted from the fault signal time series as fault signal samples, to obtain a signal data set;
[0042] S4. Establish a multi-classification support vector machine as a fault diagnosis model, and use the signal data set to train the fault diagnosis model to obtain a trained fault diagnosis model.
[0043] Furthermore, the Gram angular field transform is used to convert the time series into a two-dimensional image:
[0044] Step 1. Get the time series to be converted X = {x1, x2, ..., x t ,…,x L}, where L represents the length of the time series, scaling the time series X to a uniform scale space;
[0045] Step 2. Encode the values of the time series as the cosine of the angle, with the timestamp as the radius, and re-express the scaled time series in polar coordinates using the following equation:
[0046]
[0047] In the formula, t i is the timestamp of the time series, N is a preset constant factor, It is a time series At time ti The value of
[0048] Step 3. Define the time correlation of different time intervals by calculating the trigonometric sum or trigonometric difference between each point, and obtain the Gram angular field matrix, which is a two-dimensional image of the time series transformation.
[0049] Furthermore, extracting a feature descriptor from a two-dimensional image using a histogram of oriented gradients comprises the following steps:
[0050] 1) Grayscale the two-dimensional image;
[0051] 2) Performing global image normalization on the two-dimensional image after grayscale processing;
[0052] 3) Calculate the gradient of each pixel, including size and direction information;
[0053] 4) Divide the entire two-dimensional image into multiple blocks, each block includes multiple cell units, and calculate the gradient histogram of each cell unit respectively;
[0054] 5) For each block, concatenate the gradient histograms of all cells in the block into a vector, normalize the vector, and obtain the feature descriptor of each block;
[0055] 6) Merge the feature descriptors of each block to obtain the feature descriptor of the entire image.
[0056] Compared with the prior art, the present invention has the following beneficial effects:
[0057] (1) The Gram angle field is used to transform the sensor time series signal, and the oriented gradient histogram is used to extract the feature descriptor. This changes the way the fault signal features are extracted, amplifies the features of tiny signals, enhances the characterization ability of fault signals, and improves the accuracy and reliability of the multi-sensor fault signal classification results of gas turbines.
[0058] (2) Adding fault signals to the original signal time series can meet the needs of industrial sites and perform real-time fault signal classification from different working conditions, different fault types, different fault severities, different fault locations, etc.
[0059] (3) Signal feature extraction was performed using the Gram angular field transform and oriented gradient histogram feature descriptor extraction to improve the authenticity and accuracy of the multi-sensor fault signal classification results of the gas turbine, so that the gas turbine control system has a certain degree of automated fault diagnosis capability. BRIEF DESCRIPTION OF THE DRAWINGS
[0060] Figure 1 The flowchart of a gas turbine sensor fault diagnosis method based on visual enhancement is shown in FIG. DETAILED DESCRIPTION
[0061] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. This embodiment is implemented on the premise of the technical solution of the present invention, and a detailed implementation method and a specific operation process are given. Obviously, the described embodiment is only an embodiment of a part of the present invention, not all embodiments, and the protection scope of the present invention is not limited to the following embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work should fall within the scope of protection of the present invention.
[0062] As used herein, "one embodiment" or "embodiment" refers to a specific feature, structure or characteristic that may be included in at least one implementation of the present invention. In the description of the present invention, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units that are not listed, or may optionally include other steps or units that are inherent to these processes, methods, products or devices.
[0063] This specification provides method operation steps such as embodiments or flowcharts, but more or fewer operation steps may be included based on conventional or non-creative labor. The order of steps listed in the embodiments is only one way of executing the order of many steps and does not represent the only execution order. When the actual system or server product is executed, it can be executed in sequence or in parallel (for example, in a parallel processor or multi-threaded processing environment) according to the method shown in the embodiments or the drawings, or the execution order of steps without timing restrictions can be adjusted.
[0064] Example:
[0065] The 9F-class gas turbine is a representative gas turbine model, and therefore the embodiments of the present invention are explained by taking the sensor signal of the 9F-class gas turbine as an example.
[0066] A gas turbine sensor fault diagnosis method based on visual enhancement, such as Figure 1 As shown, the following steps are included:
[0067] S1. Collecting sensor signals of the gas turbine to obtain an original signal time series, adding different sensor faults to the original signal time series to obtain a fault signal time series;
[0068] The sensor type is a sensor with hardware redundancy, or a sensor located in the same component or the same sub-control loop. The sensors with hardware redundancy include pressure sensors, temperature sensors, and rotation speed sensors.
[0069] The gas turbine sensor signal is collected, that is, the sensor signal of the 9F-class gas turbine under normal working conditions is collected to obtain the original signal. A time series of a fixed length is intercepted on the original signal using a time window to obtain the original signal time series.
[0070] In this embodiment, the sensor data of the gas turbine is collected as the original signal (also called normal signal) source of the experimental data, and the signal data set required by this application is prepared on this basis. In other implementations, the signal data set can also be prepared by obtaining a public data set.
[0071] After obtaining the original signal time series, add step, short circuit, drift, noise, and periodic signals of different amplitudes to superimpose different sensor faults on the original signal time series, thereby obtaining a fault signal time series. In order to be closer to the real fault signal, the added sensor fault is a small fault to simulate a small-amplitude fault signal. It can be understood that an original signal time series can be superimposed with different faults to obtain multiple fault signal time series.
[0072] S2, convert the time series into a two-dimensional image using the Gram angular field transform, encode the time series values as angle cosines, and record the timestamp as the radius;
[0073] S21, obtain the time series to be converted X = {x1, x2, ..., x t ,…,x L}, where L represents the length of the time series, and the time series X is scaled to a uniform scale space [-1, 1] or [0, 1];
[0074]
[0075] or
[0076] is the value of the time series;
[0077] S22. Encode the values of the time series as the cosine of the angle, with the timestamp as the radius, and re-express the scaled time series in polar coordinates using the following equation:
[0078]
[0079] In the formula, t i is the timestamp of the time series, N is a preset constant factor used to normalize the span of the polar coordinate system. The length of the time series can be used as N. It is a time series At time t i The value of
[0080] This polar-based representation is a new way to understand time series. As time increases, the corresponding values distort between different angles on a circle, like ripples on water. The encoding graph in the above formula has two important properties. First, it is bijective, because when θ∈[0,π], cos(θ) is monotonic. Given a time series, the proposed mapping produces one and only one result in the polar coordinate system, with a unique inverse mapping. Second, in contrast to Cartesian coordinates, polar coordinates preserve absolute time relationships.
[0081] The rescaled data at different intervals have different angular bounds. [0,1] corresponds to [0,π / 2] in the cosine function, while the cosine values in the interval [-1,1] fall into the angular range [0,π]. As discussed later, they provide different information granularity for classification tasks in the Gram angle field, and the [0,1] rescaled data of the Gram angle difference field has an accurate inverse mapping.
[0082] S23. Define the time correlation of different time intervals by calculating the trigonometric sum or trigonometric difference between each point, and obtain the Gram angular field matrix, that is, the two-dimensional image of the time series conversion.
[0083] After converting the rescaled time series to polar coordinates, the angular perspective can be easily used to identify temporal correlations in different time intervals by considering the trigonometric sums / differences between each point. The Gram Sum Angular Field (GASF) and Gram Difference Angular Field (GADF) are defined as follows:
[0084]
[0085]
[0086] Where I is the unit row vector [1,1,…,1]. After transformation to polar coordinates, the time series at each time step is treated as a one-dimensional metric space. Gram angle fields have several advantages. First, they provide a way to preserve time dependencies because time increases as the position moves from the upper left corner to the lower right corner. Gram angle fields contain time dependencies because G (i,j||i-j|=k)) Represents the relative correlation of directional superposition / difference with respect to time interval k. The main diagonal G i,i is a special case when k=0, which contains the original value / angle information. However, the Gram angle field is large because the size of the Gram matrix is L×L when the length of the original time series is L. In order to reduce the size of the Gram angle field, a piecewise aggregation approximation is applied to smooth the time series while preserving the trend.
[0087] S3, using the oriented gradient histogram to extract feature descriptors in the two-dimensional image, taking the feature descriptors extracted from the original signal time series as normal signal samples, and taking the feature descriptors extracted from the fault signal time series as fault signal samples, to obtain a signal data set;
[0088] Extracting feature descriptors from a two-dimensional image using histogram of oriented gradients includes the following steps:
[0089] 1) grayscale the pseudo-color two-dimensional image obtained by Gram's differential angular field conversion;
[0090] 2) Apply an optional global image normalization equalization to perform global image normalization on the grayscaled 2D image to reduce the impact of lighting effects. Gamma power law compression can be used to calculate the square root or logarithm of each color channel;
[0091] 3) Calculate the gradient of each pixel, including size information and direction information; in this embodiment, the first-order image gradient is calculated. These capture boundaries, contours, and some texture information, while further resisting illumination changes. In other embodiments, the variant method may also include second-order image derivatives, which act as primitive strip detectors to capture useful features, such as gradient changes in the Gram angle field of step faults.
[0092] 4) Divide the entire two-dimensional image into multiple blocks, each block includes multiple cell units, and calculate the gradient histogram of each cell unit respectively;
[0093] 5) For each block, concatenate the gradient histograms of all cells in the block into a vector, normalize the vector, and obtain the feature descriptor of each block, which can also be called a feature descriptor;
[0094] Steps 4) and 5) aim to produce an encoding that is sensitive to local image content while remaining robust to noise or small changes in appearance, with the main goal of aggregating gradient information locally.
[0095] 6) Merge the feature descriptors of each block to obtain the feature descriptor of the entire image.
[0096] Cross-block normalization takes groups of local cells and contrast-normalizes their overall responses before passing them to the next stage. Normalization introduces better invariance to lighting, shadow, and edge contrast. The normalized block descriptors are called Histogram of Oriented Gradients (HOG) descriptors. Finally, HOG descriptors are collected from all blocks in a dense overlapping grid covering the detection window (the entire 2D image) into a combined feature vector.
[0097] S4, establishing a multi-classification support vector machine as a fault diagnosis model, and using the signal data set to train the fault diagnosis model to obtain a trained fault diagnosis model;
[0098] The signal data set is divided into a training set, a validation set and a test set, and the fault diagnosis model is trained based on the training set, the validation set and the test set to obtain a trained fault diagnosis model, and the classification model is trained using the training set, validated using the validation set, and tested using the test set. In this embodiment, in order to simulate the sparse fault signal in the actual operation of the gas turbine, the training set, the validation set and the test set are all unbalanced data sets, in which the ratio of normal signal samples to various fault signals is 10:3.
[0099] In this embodiment, the multi-classification support vector machine uses folding loss, and the loss function is as follows:
[0100]
[0101] In the formula, the first term is the regularization penalty, the second term is the multi-classification data loss, n represents the amount of data in the training set, and x i is the input data, y i is the label, W,b is the model parameter, φ(x i ) represents the kernel function, and C is the trade-off between the two items.
[0102] Input data x i is the HOG extraction result corresponding to each subsequence converted into a two-dimensional image in the previous step. This estimator uses stochastic gradient descent (SGD) learning to implement a regularized linear model: the loss gradient of each sample is estimated each time, and the model is continuously updated as the learning rate decreases. SGD allows small batch online learning through partial fitting methods. In order to obtain the best results with the default learning rate schedule, the input data needs to have zero mean and unit variance.
[0103] A regularizer is a penalty term added to the loss function that shrinks the model parameters to the zero vector using either the squared Euclidean norm or the absolute norm or a combination of both (Elastic Net). If a parameter update exceeds a 0.0 value due to the regularization, the update is truncated to 0.0 to allow learning of sparse models and enable online feature selection.
[0104] S5. Obtain the sensor signal to be diagnosed, perform Gram angle field transformation and feature descriptor extraction, input the extracted feature descriptor into the trained fault diagnosis model, and obtain the fault diagnosis result.
[0105] The confusion matrix can be used to visualize the multi-classification results, and the accuracy rate can be used to obtain the overall results to evaluate the classification results of the fault diagnosis model.
[0106] The sensor signals collected by the real gas turbine control system are accompanied by white noise. The feature extraction method of the fault signal should not only be free from noise interference, but also reflect the characteristics of the fault signal. However, the amplitude of the fault signal is very small under many working conditions, and it is easy to be indistinguishable from the changing trend of the normal signal. Only an effective feature extraction method can make the fault signal classification proceed smoothly.
[0107] The present application adopts the Gram angle field to transform the sensor time series signal, changes the method of extracting fault signal features, amplifies the features of tiny signals, enhances the characterization capability of fault signals, and improves the accuracy and reliability of multi-sensor fault signal classification results of gas turbines.
[0108] The present application also provides a gas turbine sensor fault diagnosis system based on visual enhancement, comprising a data acquisition module, a Gram angle field transformation module, a feature descriptor extraction module and a classification module;
[0109] The data acquisition module is used to collect sensor signals to be diagnosed;
[0110] The Gram angular field transform module is used to convert the sensor time series to be diagnosed into a two-dimensional image using the Gram angular field transform, encode the value of the time series as the angle cosine, and record the time stamp as the radius;
[0111] The feature descriptor extraction module is used to extract feature descriptors in a two-dimensional image using a histogram of oriented gradients;
[0112] The classification module is equipped with a trained classification model, which is used to input the feature descriptor of the signal to be detected into the feature model to obtain the classification result as the fault diagnosis result;
[0113] The training process of the classification model includes the following steps:
[0114] S1. Collecting sensor signals of the gas turbine to obtain an original signal time series, adding different sensor faults to the original signal time series to obtain a fault signal time series;
[0115] S2, convert the time series into a two-dimensional image using the Gram angular field transform, encode the time series values as angle cosines, and record the timestamp as the radius;
[0116] S3, using the oriented gradient histogram to extract feature descriptors in the two-dimensional image, taking the feature descriptors extracted from the original signal time series as normal signal samples, and taking the feature descriptors extracted from the fault signal time series as fault signal samples, to obtain a signal data set;
[0117] S4. Establish a multi-classification support vector machine as a fault diagnosis model, and use the signal data set to train the fault diagnosis model to obtain a trained fault diagnosis model.
[0118] The specific contents of the above-mentioned Gram angular field transformation, extracting feature descriptors and training multi-classification support vector machines have been described in a gas turbine sensor fault diagnosis based on visual enhancement, and will not be repeated here.
[0119] It should be noted that the present application can be implemented in software and / or a combination of software and hardware, for example, can be implemented using an application specific integrated circuit (ASIC), a general purpose computer or any other similar hardware device. In one embodiment, the software program of the present application can be executed by a processor to implement the steps or functions described above. Similarly, the software program of the present application (including relevant data structures) can be stored in a computer-readable recording medium, for example, a RAM memory, a magnetic or optical drive or a floppy disk and similar devices. In addition, some steps or functions of the present application can be implemented using hardware, for example, as a circuit that cooperates with a processor to perform each step or function.
[0120] In addition, a part of the present application may be applied as a computer program product, such as a computer program instruction, which, when executed by a computer, can call or provide the method and / or technical solution according to the present application through the operation of the computer. The program instruction for calling the method of the present application may be stored in a fixed or removable recording medium, and / or transmitted through a data stream in a broadcast or other signal-bearing medium, and / or stored in a working memory of a computer device that runs according to the program instruction. Here, according to an embodiment of the present application, a device is included, the device including a memory for storing computer program instructions and a processor for executing program instructions, wherein, when the computer program instruction is executed by the processor, the device is triggered to run the method and / or technical solution based on the aforementioned multiple embodiments according to the present application.
[0121] The preferred specific embodiments of the present invention are described in detail above. It should be understood that a person skilled in the art can make many modifications and changes based on the concept of the present invention without creative work. Therefore, any technical solution that can be obtained by a person skilled in the art through logical analysis, reasoning or limited experiments based on the concept of the present invention on the basis of the prior art should be within the scope of protection determined by the claims.
Claims
1. A gas turbine sensor fault diagnosis method based on visual enhancement, characterized in that: The following steps are involved: S1. Collecting sensor signals of the gas turbine to obtain an original signal time series, adding different sensor faults to the original signal time series to obtain a fault signal time series; S2, convert the time series into a two-dimensional image using the Gram angular field transform, encode the time series values as angle cosines, and record the timestamp as the radius; S3, using the oriented gradient histogram to extract feature descriptors in the two-dimensional image, taking the feature descriptors extracted from the original signal time series as normal signal samples, and taking the feature descriptors extracted from the fault signal time series as fault signal samples, to obtain a signal data set; S4, establishing a multi-classification support vector machine as a fault diagnosis model, and using the signal data set to train the fault diagnosis model to obtain a trained fault diagnosis model; S5. Obtain the sensor signal to be diagnosed, perform Gram angle field transformation and feature descriptor extraction, input the extracted feature descriptor into the trained fault diagnosis model, and obtain the fault diagnosis result.
2. A gas turbine sensor fault diagnosis method based on visual enhancement according to claim 1, characterized in that: Step S2 is specifically as follows: S21, obtain the time series to be converted x = {x1, x2, ..., x t ,…,x L }, where L represents the length of the time series, scaling the time series X to a uniform scale space; S22. Encode the values of the time series as the cosine of the angle, with the timestamp as the radius, and re-express the scaled time series in polar coordinates using the following equation: Where, t i is the timestamp of the time series, N is a preset constant factor, It is a time series At time t i The value of S23. Define the time correlation of different time intervals by calculating the trigonometric sum or trigonometric difference between each point, and obtain the Gram angular field matrix, that is, the two-dimensional image of the time series conversion.
3. A gas turbine sensor fault diagnosis method based on visual enhancement according to claim 1, characterized in that: In step S3, extracting feature descriptors from a two-dimensional image using a histogram of oriented gradients includes the following steps: 1) Grayscale the two-dimensional image; 2) Performing global image normalization on the two-dimensional image after grayscale processing; 3) Calculate the gradient of each pixel, including size and direction information; 4) Divide the entire two-dimensional image into multiple blocks, each block includes multiple cell units, and calculate the gradient histogram of each cell unit respectively; 5) For each block, concatenate the gradient histograms of all cells in the block into a vector, normalize the vector, and obtain the feature descriptor of each block; 6) Merge the feature descriptors of each block to obtain the feature descriptor of the entire image.
4. A gas turbine sensor fault diagnosis method based on visual enhancement according to claim 1, characterized in that: The signal data set is divided into a training set, a validation set and a test set, and the fault diagnosis model is trained based on the training set, the validation set and the test set to obtain a trained fault diagnosis model.
5. A gas turbine sensor fault diagnosis method based on visual enhancement according to claim 4, characterized in that: The training set, validation set and test set are unbalanced data sets, in which the ratio of normal signal samples to various types of fault signals is 10:
3.
6. A gas turbine sensor fault diagnosis method based on visual enhancement according to claim 4, characterized in that: The loss function of the multi-classification support vector machine is as follows: In the formula, the first term is the regularization penalty, the second term is the multi-classification data loss, n represents the amount of data in the training set, and x i is the input data, y i is the label, W,b is the model parameter, φ(x i ) represents the kernel function, and C is the trade-off between the two items.
7. A gas turbine sensor fault diagnosis method based on visual enhancement according to claim 1, characterized in that: In step S1, adding different sensor faults to the original signal time series specifically includes: adding steps, short circuits, drifts, noises, and periodic signals of different amplitudes to the sensor signal.
8. A gas turbine sensor fault diagnosis system based on visual enhancement, characterized in that: It includes a data acquisition module, a Gram angular field transformation module, a feature descriptor extraction module and a classification module; The data acquisition module is used to collect sensor signals to be diagnosed; The Gram angular field transform module is used to convert the sensor time series to be diagnosed into a two-dimensional image using the Gram angular field transform, encode the value of the time series as the angle cosine, and record the time stamp as the radius; The feature descriptor extraction module is used to extract feature descriptors in a two-dimensional image using a histogram of oriented gradients; The classification module is equipped with a trained classification model, which is used to input the feature descriptor of the signal to be detected into the feature model to obtain the classification result as the fault diagnosis result; The training process of the classification model includes the following steps: S1. Collecting sensor signals of the gas turbine to obtain an original signal time series, adding different sensor faults to the original signal time series to obtain a fault signal time series; S2, convert the time series into a two-dimensional image using the Gram angular field transform, encode the time series values as angle cosines, and record the timestamp as the radius; S3, using the oriented gradient histogram to extract feature descriptors in the two-dimensional image, taking the feature descriptors extracted from the original signal time series as normal signal samples, and taking the feature descriptors extracted from the fault signal time series as fault signal samples, to obtain a signal data set; S4. Establish a multi-classification support vector machine as a fault diagnosis model, and use the signal data set to train the fault diagnosis model to obtain a trained fault diagnosis model.
9. A gas turbine sensor fault diagnosis system based on visual enhancement according to claim 8, characterized in that: The Gram angular field transform is used to convert the time series into a two-dimensional image: Step 1, get the time series to be converted X = {x1, X2, ..., X t ,…,X L }, where L represents the length of the time series, scaling the time series X to a uniform scale space; Step 2. Encode the values of the time series as the cosine of the angle, with the timestamp as the radius, and re-express the scaled time series in polar coordinates using the following equation: Where, t i is the timestamp of the time series, N is a preset constant factor, It is a time series At time t i The value of Step 3. Define the time correlation of different time intervals by calculating the trigonometric sum or trigonometric difference between each point, and obtain the Gram angular field matrix, that is, the two-dimensional image of the time series transformation.
10. A gas turbine sensor fault diagnosis system based on visual enhancement according to claim 8, characterized in that: Extracting feature descriptors from a two-dimensional image using histogram of oriented gradients includes the following steps: 1) Grayscale the two-dimensional image; 2) Performing global image normalization on the two-dimensional image after grayscale processing; 3) Calculate the gradient of each pixel, including size and direction information; 4) Divide the entire two-dimensional image into multiple blocks, each block includes multiple cell units, and calculate the gradient histogram of each cell unit respectively; 5) For each block, concatenate the gradient histograms of all cells in the block into a vector, normalize the vector, and obtain the feature descriptor of each block; 6) Merge the feature descriptors of each block to obtain the feature descriptor of the entire image.
