Fault diagnosis method and system for a capacitor thermal management system of a tram

Abstract

The present application discloses a fault diagnosis method and system for a capacitor thermal management system of a tram. The fault diagnosis method includes: a data acquisition step: collecting operation data of the supercapacitor thermal management system of the tram and constructing an operation data set according to the operation data; a data preprocessing step: processing the operation data set to obtain the total error value of the operation data set; a BP neural network optimization step: calculating the learning rate of the BP neural network according to the total error value, and optimizing the BP neural network according to the learning rate of the BP neural network and the additional momentum factor. After optimization, according to the weights of the first layer and the second layer of the optimized BP neural network, calculating the preliminary fault diagnosis result output by the optimized BP neural network; a fault type and fault probability prediction result obtaining step: the KNN&P fault diagnosis and probability prediction model outputs the fault type and the fault probability prediction result according to the preliminary fault diagnosis result.

Description

Technical Field

[0001] This application relates to the technical field of the detection of the thermal management system of rail vehicles, and particularly relates to a fault diagnosis method and system for the thermal management system of a tram capacitor. Background Art

[0002] Trams are mainly powered by overhead lines. On roads without overhead lines or at intersections with a large number of over-height vehicles, on-vehicle energy storage is used. As the main energy storage component, the supercapacitor module has become a key part of trams. The stability, reliability, and timely fault diagnosis system of the thermal management system that ensures the normal operation of supercapacitors are all crucial. The daily maintenance, potential hazard investigation, and fault diagnosis of its thermal management system are important means to ensure the reliability of train operation.

[0003] The fault situation of the supercapacitor thermal management system is subject to on-site notifications. Fault handling lags far behind the occurrence of faults and potential faults, and has a serious impact on the normal train operation order after a fault. Fault investigation methods often rely on technicians with rich experience to climb to the top of the vehicle on-site for empirical analysis and investigation based on fault phenomena, and require complex unpacking operations. The investigation time is usually long and difficult, and some fault phenomena are not obvious, bringing great difficulties to on-site investigators. For the traditional fault diagnosis model of the thermal management system, there are characteristics such as complex model, long training time, low accuracy, and inability to calculate the fault probability. Therefore, in the process of detecting the supercapacitor thermal management system of trams, how to achieve fault diagnosis based on the BP-KNN&P algorithm model and the inability of the fault diagnosis model to calculate the fault probability has become an urgent problem to be solved. Summary of the Invention

[0004] The embodiments of this application provide a fault diagnosis method and system for the thermal management system of a tram capacitor, so as to at least solve problems such as complex model, long model training time, low fault diagnosis accuracy, inability to achieve fault diagnosis based on the BP-KNN&P algorithm model, inability of the fault diagnosis model to calculate the fault probability, and high difficulty in fault investigation and solution through the present invention.

[0005] The present invention provides a fault diagnosis method, including:

[0006] Data acquisition step: Collect the operation data of the supercapacitor thermal management system of a tram, and construct an operation data set according to the operation data;

[0007] Data preprocessing step: Process the operation data set to obtain the total error value of the operation data set;

[0008] Steps for optimizing the BP neural network: Calculate the learning rate of the BP neural network according to the total error value. After optimizing the BP neural network according to the learning rate of the BP neural network and the additional momentum factor, calculate the preliminary result of fault diagnosis output by the optimized BP neural network according to the weight values of the first layer and the second layer of the optimized BP neural network.

[0009] Steps for obtaining the fault type and fault probability prediction result: The KNN&P fault diagnosis and probability prediction model outputs the fault type and fault probability prediction result according to the preliminary result of fault diagnosis.

[0010] The above-mentioned fault diagnosis method, wherein the data preprocessing step includes:

[0011] After normalizing the operation data set, obtain the fault label by corresponding the normalized operation data set with the fault label, and divide the normalized operation data set into an operation data training set and an operation data test set.

[0012] The above-mentioned fault diagnosis method, wherein the preprocessing step further includes:

[0013] Initialize the neural network weight values of the BP neural network with a Xavier uniform distribution to obtain the weight values of the first layer and the second layer of the BP neural network.

[0014] Form an activation function according to the normalized operation data set, the fault label, the initialized weight value of the first layer, and the weight value of the second layer through the SoftSign function.

[0015] Construct an error function according to the activation function, and calculate the total error value of the operation data training set through the error function.

[0016] The above-mentioned fault diagnosis method, wherein the BP neural network optimization step includes:

[0017] Calculate the additional momentum factor and the learning rate of the BP neural network according to the total error value.

[0018] Train the BP network according to the learning rate of the BP neural network and the additional momentum factor to obtain the change amounts of the weight values of the first layer and the second layer of the optimized BP neural network.

[0019] The above-mentioned fault diagnosis method, wherein the BP neural network optimization step further includes:

[0020] Based on the first-layer weight change amount and the second-layer weight change amount, the first-layer weight and the second-layer weight of the optimized BP neural network are calculated through the backpropagation method;

[0021] Based on the first-layer weight and the second-layer weight, the preliminary fault diagnosis result is calculated.

[0022] The above-mentioned fault diagnosis method, wherein the step of obtaining the fault type and the fault probability prediction result includes:

[0023] Construct the KNN&P fault diagnosis and probability prediction model based on the nearest neighbor KNN&P algorithm with K = 1, test the KNN&P fault diagnosis and probability prediction model through the operation data test set, and obtain the fault type and the fault probability prediction result through the KNN&P fault diagnosis and probability prediction model after testing of the sample set.

[0024] The above-mentioned fault diagnosis method, wherein the step of obtaining the fault type and the fault probability prediction result includes:

[0025] According to the preliminary fault diagnosis result, obtain the distance from the preliminary fault diagnosis result to the fault label through the distance measurement method of Euclidean distance;

[0026] According to the nearest neighbor KNN&P algorithm with K = 1, take the fault label value closest to the preliminary fault diagnosis result.

[0027] The present invention also provides a fault diagnosis system, which is applicable to the above-mentioned fault diagnosis method, and the fault diagnosis system includes:

[0028] Data acquisition unit: Collect the operation data of the tram supercapacitor thermal management system through sensors, construct an operation data set according to the operation data, and store the operation data in a cloud server through an Internet of Things platform;

[0029] Data preprocessing unit: In the data training and analysis platform, process the operation data set to obtain the total error value of the operation data set;

[0030] BP neural network optimization unit: Calculate the learning rate of the BP neural network according to the total error value, optimize the BP neural network according to the learning rate of the BP neural network and the additional momentum factor, and calculate the preliminary fault diagnosis result output by the optimized BP neural network according to the first-layer weight and the second-layer weight of the optimized BP neural network;

[0031] Fault type and fault probability prediction result acquisition unit: The KNN&P fault diagnosis and probability prediction model outputs the fault type and fault probability prediction result according to the preliminary fault diagnosis result, and synchronizes the fault type and the fault probability prediction result to the after-sales service platform.

[0032] The above-mentioned fault diagnosis system, wherein the operation data set includes the operation data of the pipeline part of the thermal management system, the operation data of the electrical part of the thermal management system, the data of the third supercapacitor module, the operation state data of the devices of the thermal management system, and the working mode feedback data of the devices of the thermal management system. The data acquisition unit acquires the operation state data of the devices of the thermal management system and the working mode feedback data through the IO of the thermal management system controller.

[0033] The above-mentioned fault diagnosis system, wherein the data acquisition unit further includes:

[0034] Temperature sensors and pressure sensors, which acquire the operation data of the pipeline part of the thermal management system through the temperature sensors and pressure sensors;

[0035] Current sensors and voltage sensors, which acquire the operation data of the electrical part of the thermal management system through the current sensors and voltage sensors;

[0036] Humidity sensors and the temperature sensors, which acquire the data of the first supercapacitor module through the humidity sensors and the temperature sensors;

[0037] Wherein, the data acquisition unit acquires the data of the second supercapacitor module through the vehicle CAN communication, and analyzes the data of the first supercapacitor module and the data of the second supercapacitor module to obtain the third supercapacitor module.

[0038] Compared with the related technologies, a fault diagnosis method and system for a tram capacitor thermal management system proposed by the present invention introduce the concept of PHM (Prognostics and Health Management) into the tram supercapacitor thermal management system, solving the problem that there is no related technology for the fault diagnosis method and device of the supercapacitor module thermal management system at present; optimize the traditional BP neural network that falls into local minimum with additional momentum, and improve the learning rate and stability of the traditional BP neural network to an adaptive learning rate. At the same time, select a weight initialization method and activation function suitable for the tram supercapacitor thermal management system, so that the improved traditional BP neural network is adapted to the fault diagnosis of the tram supercapacitor thermal management system; to adapt to the fault diagnosis of the tram supercapacitor thermal management system, the present invention improves the traditional KNN classification algorithm to obtain a KNN&P fault classification and probability prediction model, and classifies the neural network diagnosis results probabilistically through the KNN&P fault classification and probability prediction model, diagnosing the probability of each fault, and solving the problem that the current fault diagnosis model cannot calculate the fault probability; to adapt to the fault diagnosis of the tram supercapacitor thermal management system, combine the improved BP neural network with the improved KNN classification algorithm to obtain a KNN&P fault classification and probability prediction model, realizing timely fault diagnosis of the tram supercapacitor thermal management system and improving the diagnosis accuracy; the present invention combines the Internet of Things with the fault diagnosis of the tram supercapacitor thermal management system, and proposes a remote real-time monitoring strategy, which supports real-time two-way data transmission between the fault diagnosis system end and the WEB end and the mobile end, improving the efficiency of maintenance personnel in monitoring the operation status and faults of the tram supercapacitor thermal management system and program updating and other tasks.

[0039] Details of one or more embodiments of the present application are set forth in the following drawings and description to make other features, objects, and advantages of the present application more concise and understandable. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] The drawings described herein are used to provide a further understanding of the present application, and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation of the present application. In the drawings:

[0041] Figure 1 is a flowchart of a fault diagnosis method for a tram capacitor thermal management system according to an embodiment of the present application;

[0042] Figure 2 is a layout diagram of a fault diagnosis device for a supercapacitor module thermal management system according to an embodiment of the present application;

[0043] Figure 3 is a fault diagnosis parameter diagram of a thermal management system according to an embodiment of the present application;

[0044] Figure 4 It is a framework diagram of the KNN&P neural network model according to an embodiment of the present application;

[0045] Figure 5 It is a flowchart of the KNN&P fault diagnosis algorithm model according to an embodiment of the present application;

[0046] Figure 6 It is a schematic structural diagram of the fault diagnosis system of the tram capacitor thermal management system of the present invention;

[0047] Figure 7 It is a framework diagram of the data flow direction according to an embodiment of the present application.

[0048] Among them, the reference numerals are:

[0049] Data acquisition unit: 51;

[0050] Data preprocessing unit: 52;

[0051] BP neural network optimization unit: 53;

[0052] Fault type and fault probability prediction result obtaining unit: 54. Specific implementation manners

[0053] In order to make the purpose, technical solutions and advantages of the present application clearer and more understandable, the present application will be described and explained below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application. Based on the embodiments provided by the present application, all other embodiments obtained by those of ordinary skill in the art without making creative efforts belong to the scope of protection of the present application.

[0054] Obviously, the accompanying drawings in the following description are only some examples or embodiments of the present application. For those of ordinary skill in the art, without making creative efforts, the present application can also be applied to other similar scenarios based on these drawings. In addition, it can also be understood that although the efforts made in this development process may be complex and lengthy, for those of ordinary skill in the art related to the disclosure content of the present application, some design, manufacturing or production changes based on the technical content disclosed in the present application are only conventional technical means and should not be understood that the disclosure content of the present application is insufficient.

[0055] References to "embodiments" in this application mean that specific features, structures, or characteristics described in connection with the embodiments can be included in at least one embodiment of this application. The phrase appears in various places in the specification and does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment mutually exclusive with other embodiments. Those of ordinary skill in the art will explicitly and implicitly understand that the embodiments described in this application can be combined with other embodiments without conflict.

[0056] Unless otherwise defined, technical terms or scientific terms involved in this application should have the ordinary meaning understood by those with ordinary skills in the technical field to which this application belongs. The words "a", "one", "kind", "the" and similar words involved in this application do not indicate a limitation in quantity and can represent a singular or plural number. The terms "including", "comprising", "having" and any variations thereof involved in this application are intended to cover non-exclusive inclusion; for example, a process, method, system, product or device that includes a series of steps or modules (units) is not limited to the listed steps or units, but may further include unlisted steps or units, or may further include other steps or units inherent to these processes, methods, products or devices. The terms "connected", "coupled" and similar words involved in this application are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. The term "plurality" involved in this application means two or more. "And / or" describes the association relationship of associated objects and indicates that three relationships may exist. For example, "A and / or B" may mean: A exists alone, A and B exist simultaneously, and B exists alone. The character " / " generally indicates that the associated objects before and after are in an "or" relationship. The terms "first", "second", "third", etc. involved in this application are only used to distinguish similar objects and do not represent a specific order for the objects.

[0057] The present invention provides a fault diagnosis method and system for a capacitor thermal management system of a tram. The present invention establishes a KNN&P neural network fault diagnosis and probability prediction model that is concise, has a short training time, high fault diagnosis accuracy, and can predict the probability of faults. A variety of optimization algorithms are used to optimize the neural network fault diagnosis model. The fault training set data is used for the training of the KNN&P neural network fault diagnosis and probability prediction model. After obtaining the optimized KNN&P neural network fault diagnosis and probability prediction model, the BP-KNN&P neural network fault diagnosis and probability prediction model is applied to the fault diagnosis of the supercapacitor thermal management system of the tram, providing fault diagnosis information for the staff in a timely manner, reducing the difficulty of fault troubleshooting and solution, and greatly reducing the impact on the train operation order.

[0058] The present invention will be described below in conjunction with specific embodiments.

[0059] Embodiment 1

[0060] This embodiment also provides a fault diagnosis method for the capacitor thermal management system of a tram. Please refer to Figures 1 to 5 , Figure 1 which is the flowchart of the fault diagnosis method for the capacitor thermal management system according to the embodiment of the present application; Figure 2 which is the layout diagram of the fault diagnosis device for the supercapacitor module thermal management system according to the embodiment of the present application; Figure 3 which is the fault diagnosis parameter diagram of the thermal management system according to the embodiment of the present application; Figure 4 which is the framework diagram of the KNN&P neural network model according to the embodiment of the present application; Figure 5 which is the flowchart of the KNN&P fault diagnosis algorithm model according to the embodiment of the present application. As Figures 1 to 5 shown, the fault diagnosis method includes:

[0061] Data acquisition step S1: Acquire the operation data of the supercapacitor thermal management system of the tram, and construct an operation data set according to the operation data;

[0062] Data preprocessing step S2: Process the operation data set to obtain the total error value of the operation data set;

[0063] BP neural network optimization step S3: Calculate the learning rate of the BP neural network according to the total error value, optimize the BP neural network according to the learning rate of the BP neural network and the additional momentum factor, and calculate the preliminary fault diagnosis result output by the optimized BP neural network according to the first-layer weight and the second-layer weight of the optimized BP neural network;

[0064] Fault type and fault probability prediction result obtaining step: The KNN&P fault diagnosis and probability prediction model outputs the fault type and fault probability prediction result according to the preliminary fault diagnosis result.

[0065] In the embodiment, the data preprocessing step S1 includes:

[0066] After normalizing the operation data set, corresponding the normalized operation data set with the fault label to obtain the fault label, and dividing the normalized operation data set into an operation data training set and an operation data test set;

[0067] Initialize the neural network weights of the BP neural network with a Xavier uniform distribution to obtain the first-layer weight and the second-layer weight of the BP neural network;

[0068] An activation function is formed by the SoftSign function based on the normalized operating data set, the fault label, the initialized first-layer weight, and the second-layer weight.

[0069] An error function is constructed based on the activation function, and the total error value of the operating data training set is calculated through the error function.

[0070] In a specific implementation, data under normal conditions and various fault conditions of the thermal management system are collected respectively to form an overall operating data set as X[t, n]; the historical operating data of the tram thermal management system per unit time is in the form of a vector X(t) and is stored in a cloud server through the Internet of Things, where x n is a single data feature data, and the data content is as Figure 3 shown, where t is the data running time, and the operating data set is a two-dimensional data matrix X[t, n].

[0071] X(t) = [x1(t), x2(t), …, x n (t)] T (1)

[0072] After normalizing the operating data set, the normalized operating data set is corresponded with the fault label to obtain the fault label, and the normalized operating data set is divided into an operating data training set and an operating data test set; specifically, the original historical operating data set per unit time in formula (1) is normalized into a normalized historical operating data per unit time with a mean of 0 and a variance of 1 by formula (2), and after normalization, the data set is randomly assigned. 80% of the data per unit time in the data set forms the operating data training set X t [t, n], and 20% of the data per unit time forms the test set X e [t, n]. The operating data test set is used to detect the training results and accuracy of the KNN&P fault diagnosis and probability prediction model.

[0073]

[0074] Where σ(x) is the mean and standard deviation of all sample data of the single data feature data x i ;

[0075] The normalized operating data set is corresponded with the fault label to obtain the fault label; specifically, the normalized historical operating data per unit time in formula (1) corresponds to the fault label T(t) to form the historical operating data per unit time and the classification T(t) data shown in formula (3).

[0076] T(t) = T tk = {[y 11, y 12 , ..., y 1k ,[y 21 , y 22 , ..., y 2k ,[y t1 , y t2 , ..., y tk} (3)

[0078] Initialize the neural network weights of the BP neural network with a Xavier uniform distribution to obtain the weights of the first layer and the second layer of the BP neural network; specifically, like full same-parameter initialization and all-zero initialization, neither can break the parameter symmetry of the neural network and correct training results cannot be obtained. Using random value initialization can well break the symmetry of the neural network, but whether using too large or too small random values for initialization, the finally converged loss value is relatively large. Therefore, the present invention introduces Xavier uniform distribution initialization to initialize the neural network weights and obtain the weights of the first layer and the second layer of the initialized neural network. The weights of the first layer and the second layer of the initialized neural network are as follows:

[0079]

[0080] where V ij and W ij are the weights of the first layer and the second layer of the neural network, as shown in Equation (4);

[0081] Form an activation function through the SoftSign function according to the normalized operation data set, fault labels, and the initialized weights of the first layer and the second layer; specifically, Softsign is centered at 0, which improves the operation efficiency. For the characteristics of a large number of fault diagnosis feature variables, inconsistent feature categories, and a large sample size of tram supercapacitors, if the Sigmoid and Tanh activation functions are selected for the activation function, it is easy to cause gradient explosion and slow model training speed. Therefore, the activation functions of the hidden layer neurons and the output layer both adopt the SoftSign function, as shown in Equation (5):

[0082]

[0083] where x i is the operation data, and f1(x), f2(x) are the activation functions;

[0084] Construct an error function based on the activation function, and calculate the total error value of the operation data training set through the error function; specifically, the error function E is the total error value of the entire data training set X t [t, n], as shown in Equation (6).

[0085]

[0086] In an embodiment, the BP neural network optimization step S2 includes:

[0087] Calculating an additional momentum factor and the learning rate of the BP neural network based on the total error value;

[0088] Training the BP network according to the learning rate of the BP neural network and the additional momentum factor to obtain the change amount of the weights of the first layer and the change amount of the weights of the second layer of the optimized BP neural network;

[0089] Calculating the weights of the first layer and the weights of the second layer of the optimized BP neural network by backpropagation according to the change amount of the weights of the first layer and the change amount of the weights of the second layer;

[0090] Calculating a preliminary result of the fault diagnosis according to the weights of the first layer and the weights of the second layer.

[0091] In a specific implementation, an additional momentum factor and the learning rate of the BP neural network are calculated according to the total error value; specifically, for the training set X t [t, n], using a locally optimal neural network, learning rate, and stability-improved backpropagation to update V jk , w jk weights to solve the problem that the training process may be trapped in local minima and the training speed is slow. The BP algorithm itself is an excellent local search algorithm. The BP algorithm will converge to different local minima. The classical BP neural network backpropagation weight update method is not good. Now, the additional momentum method is used to make the network not only consider the effect of the error on the gradient when correcting its weights, but also consider the influence of the change trend on the error surface. Its function is like a low-pass filter, which allows the network to ignore the small change characteristics on the network. The weight adjustment formula with an additional momentum factor is:

[0092]

[0093] Regarding the contradiction between the learning rate and stability, the gradient algorithm requires a small learning rate for stable learning, so usually the convergence speed of the learning process is slow. If the learning rate is too large, it is prone to oscillation and difficult to converge. Now, the learning rate is changed from a fixed η to the form shown in formula (8), and the network automatically adjusts the learning rate during the training process. When the weight correction conforms to the direction of error reduction, a quantity can be added to it; when the weight correction conforms to the direction of error increase, the quantity added to it can be reduced;

[0094]

[0095] Train the BP network according to the learning rate and additional momentum factor of the BP neural network to obtain the weight change of the first layer and the weight change of the second layer of the optimized BP neural network; specifically, the formulas for calculating the weight change of the first layer of the neural network and the weight change of the second layer of the neural network are as follows:

[0096]

[0097]

[0098] Calculate the new weights of the first layer and the second layer of the neural network through the backpropagation method. The formulas for calculating the new weights of the first layer and the new weights of the second layer are as follows:

[0099] v ij = v ij + Δv ij (m), w jk = w jk + Δw jk (m); (11)

[0100] Where the value of m is the number of backpropagation times, and the use of additional momentum may cross these minima;

[0101] According to the new weights of the first layer and the second layer of the neural network, calculate the preliminary result of fault diagnosis; specifically, on the basis of the backpropagation method, add a value proportional to the previous weight change to the change of each weight, and generate a new weight change according to the backpropagation method. The first-level output of BP-KNN&P is O(t), as shown in Equation (12).

[0102]

[0103] In the embodiment, the steps S4 for obtaining the fault type and fault probability prediction result include:

[0104] Construct the KNN&P fault diagnosis and probability prediction model based on the nearest neighbor KNN&P algorithm with K = 1, test the KNN&P fault diagnosis and probability prediction model through the operation data test set, and obtain the fault type and fault probability prediction result through the KNN&P fault diagnosis and probability prediction model after testing the sample set;

[0105] According to the preliminary result of fault diagnosis, obtain the distance from the preliminary result of fault diagnosis to the fault label through the distance measurement method of Euclidean distance;

[0106] According to the nearest neighbor KNN&P algorithm with K = 1, take the fault label value closest to the preliminary result of fault diagnosis.

[0107] In specific implementation, a KNN&P fault diagnosis and probability prediction model is constructed based on the nearest neighbor KNN&P algorithm with K = 1, and the KNN&P fault diagnosis and probability prediction model is tested through a running data test set; specifically, the full name of KNN is K Nearest Neighbors, which are the K nearest neighbors, and the value of K is crucial. In the present invention, the nearest neighbor KNN&P algorithm with K = 1 is used to complete the improvement of the KNN&P algorithm; the improved KNN&P algorithm is applied to the fault diagnosis model of the thermal management system to determine the fault category and predict the probability P of each fault category, so it is improved to the KNN&P fault diagnosis and probability prediction model.

[0108] According to the preliminary fault diagnosis result, the distance from the preliminary fault diagnosis result to the fault label is obtained through the distance measurement method of Euclidean distance; according to the nearest neighbor KNN&P algorithm with K = 1, the fault label value closest to the preliminary fault diagnosis result is taken; specifically, first calculate the spatial distance L t (i):

[0109]

[0110] In formula (13), i is the type of fault label; then, according to the distance measurement method of Euclidean distance, find the fault label T tk corresponding to the primary output O tk in the KNN&P primary output O tk with the closest label value.

[0111] The KNN&P fault diagnosis and probability prediction model outputs the fault type and fault probability prediction result according to the preliminary fault diagnosis result; specifically, select the running data test set X e [t, n], detect the training result of the KNN&P fault diagnosis and probability prediction model. After the detection result is qualified, the KNN&P fault diagnosis and probability prediction model outputs the fault type and fault probability prediction result according to the preliminary fault diagnosis result, where the fault category is Q1 and the fault probability, that is, the fault probability prediction, is Q2. Q1 and Q2 are shown as follows:

[0112] Q1 = argmin(L t (i)) (14)

[0113]

[0114] where c is the total number of fault label categories, and U = (u1, u2,..., u i ) is the weight value of the distance between the predicted value O(t) and the label value T(t).

[0115] Figure 4The input X in the [system] stores historical operation data per unit time in the form of a vector X(t), and the storage format is as shown in Equation (1), and the data content is as Figure 3 shown; Figure 4 The part of the dashed box in the [system] is the improved BP neural network described above, and O e is the preliminary result of the neural network fault diagnosis, and KNN&P is the improved KNN&P fault diagnosis and probability prediction model described above, and Q g is the output fault type and probability prediction of the KNN&P model.

[0116] Embodiment 2

[0117] This embodiment also provides a fault diagnosis method flow step for a capacitor thermal management system. Please refer to Figure 5 , Figure 5 which is the flow chart of the KNN&P fault diagnosis algorithm model according to the embodiment of the present application.

[0118] Step 1: Train the improved BP neural network. If the training meets the training expectations, go to Step 2; if the training does not meet the training expectations, go back to Step 1 and train the improved BP neural network again;

[0119] Step 2: After the training meets the training expectations, the trained improved BP neural network performs fault diagnosis on the operation data set;

[0120] Step 3: The KNN&P fault classification and probability prediction model performs fault classification and fault probability prediction according to the fault diagnosis result;

[0121] Step 4: Select the operation data test set X e [t, n], and determine whether the fault diagnosis of the KNN&P fault classification and probability prediction model is qualified. If the fault diagnosis is qualified, go to Step 5; if the fault diagnosis is unqualified, adjust U=(u1, u2,..., u i ) according to the engineering experience value and go back to Step 3;

[0122] Step 5: After the fault diagnosis is qualified, the KNN&P fault classification and probability prediction model outputs the fault category and the fault probability prediction result;

[0123] Step 6: End.

[0124] Embodiment 3

[0125] This embodiment also provides a fault diagnosis system for a tram capacitor thermal management system. Figure 6 is the structural schematic diagram of the fault diagnosis system for the capacitor thermal management system of the present invention; Figure 7 is the data flow direction framework diagram according to the embodiment of the present application. As Figures 6 to 7As shown, the fault diagnosis system of the invention is applicable to the above-mentioned fault diagnosis method. The fault diagnosis system includes:

[0126] Data acquisition unit 51: Collect the operation data of the tram supercapacitor thermal management system through sensors, construct an operation data set according to the operation data, and store the operation data in the cloud server through the Internet of Things platform;

[0127] Data preprocessing unit 52: In the data training and analysis platform, process the operation data set to obtain the total error value of the operation data set;

[0128] BP neural network optimization unit 53: Calculate the learning rate of the BP neural network according to the total error value, optimize the BP neural network according to the learning rate of the BP neural network and the additional momentum factor, and calculate the preliminary fault diagnosis result output by the optimized BP neural network according to the first-layer weight and the second-layer weight of the optimized BP neural network;

[0129] Fault type and fault probability prediction result obtaining unit 54: The KNN&P fault diagnosis and probability prediction model outputs the fault type and the fault probability prediction result according to the preliminary fault diagnosis result, and synchronizes the fault type and the fault probability prediction result to the after-sales service platform.

[0130] In the embodiment, the operation data set includes the operation data of the pipeline part of the thermal management system, the operation data of the electrical part of the thermal management system, the data of the third supercapacitor module, the operation state data of the devices of the thermal management system, and the working mode feedback data of the devices of the thermal management system. The data acquisition unit collects the operation state data of the devices of the thermal management system and the working mode feedback data of the devices of the thermal management system through the IO of the thermal management system controller.

[0131] In the embodiment, the data acquisition unit 51 further includes:

[0132] Temperature sensor and pressure sensor, collect the operation data of the pipeline part of the thermal management system through the temperature sensor and the pressure sensor;

[0133] Current sensor and voltage sensor, collect the operation data of the electrical part of the thermal management system through the current sensor and the voltage sensor;

[0134] Humidity sensor and the temperature sensor, collect the data of the first supercapacitor module through the humidity sensor and the temperature sensor;

[0135] Wherein, the data acquisition unit obtains the data of the second supercapacitor module through the vehicle CAN communication, and analyzes the data of the first supercapacitor module and the data of the second supercapacitor module to obtain the third supercapacitor module.

[0136] In specific implementation, in the thermal management pipeline system, temperature sensors and pressure sensors are added at the suction port and exhaust port of the compressor of the heat exchange system, and temperature sensors are added at the evaporator and condenser to detect the real-time situation of the pipeline system. Temperature sensors and refrigerant pressure sensors are set in the pipeline part of the thermal management system; current sensors are added at the electrical and control part, the compressor, the ventilator, and the condensing fan, and a system voltage sensor is added to the total system to detect the operating state of the electrical components; the controller IO feeds back the operating state and working mode of each device and detects the coupling situation of each part of the electrical system; a temperature and humidity sensor is added at the supercapacitor module to detect the real-time state of the module;

[0137] Fault diagnosis-related sensors are installed on the pipeline system part, the electrical system part, and the supercapacitor module part of the supercapacitor thermal management system. The sensors collect the main operation data of the pipeline system part, the electrical system part, and the supercapacitor module part of the supercapacitor thermal management system; the working data of the supercapacitor is obtained through the vehicle CAN communication; the controller IO of the thermal management system collects the movement situation and working mode feedback of the devices of the thermal management system. The data content and sources, etc., are as Figure 3 shown.

[0138] In the embodiment, the fault type and fault probability prediction result obtaining unit 54 includes:

[0139] The KNN&P fault diagnosis and probability prediction model outputs the fault type and fault probability prediction result according to the preliminary fault diagnosis result, and synchronizes the fault type and the fault probability prediction result to the after-sales service platform.

[0140] In specific implementation, after the KNN&P fault diagnosis and probability prediction model outputs the fault type and fault probability prediction result, the fault type and the fault probability prediction result are synchronized to the after-sales service platform; specifically, the Internet of Things data platform displays the parameter details, the fault diagnosis and probability prediction situation, and docks with the after-sales service platform, and the after-sales personnel handle the problems in time according to the fault diagnosis situation.

[0141] In summary, at present, in the rail vehicle industry, the detection of the operating status, maintenance and update of the supercapacitor thermal management system can only be completed on-site. Once the supercapacitor thermal management system needs to be repaired, users and manufacturers must send people to the site, which not only wastes time but also increases the enterprise operation and maintenance costs. A fault diagnosis method and system for a capacitor thermal management system proposed by the present invention can solve this problem. By building a set of fault diagnosis system combined with the Internet of Things platform, the present invention enables the detection personnel to remotely monitor the operating status of the supercapacitor thermal management system and issue a fault report in a timely manner according to the detection results, accurately locate the fault location, type and probability, and improve work efficiency. At the same time, compared with the existing fault diagnosis algorithm model technology, the KNN&P type algorithm model has better technical effects and can realize fast fault diagnosis and prediction of the probability of new faults.

[0142] The above embodiments only represent several implementation manners of the present application, and the description thereof is relatively specific and detailed, but it should not be construed as a limitation on the scope of the invention patent. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present application, several modifications and improvements can still be made, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the protection scope of the appended claims.

Claims

1. A fault diagnosis method, characterized in that Applied to the fault diagnosis of the supercapacitor thermal management system of a tram, the fault diagnosis method includes: Data acquisition step: Collect the operation data of the supercapacitor thermal management system of the tram, and construct an operation data set according to the operation data; Data preprocessing step: Process the operation data set to obtain the total error value of the operation data set; BP neural network optimization step: Calculate the learning rate of the BP neural network according to the total error value. According to the learning rate of the BP neural network and the additional momentum factor, after optimizing the BP neural network, calculate the preliminary fault diagnosis result output by the optimized BP neural network according to the first-layer weight and the second-layer weight of the optimized BP neural network; Fault type and fault probability prediction result obtaining step: The KNN&P fault diagnosis and probability prediction model outputs the fault type and the fault probability prediction result according to the preliminary fault diagnosis result; Specifically, the KNN&P fault diagnosis and probability prediction model is constructed based on the nearest neighbor KNN&P algorithm with K = 1, and the KNN&P fault diagnosis and probability prediction model is tested by the operation data test set, and the sample set obtains the fault type and the fault probability prediction result through the tested KNN&P fault diagnosis and probability prediction model.

2. The fault diagnosis method according to claim 1, characterized in that, The data preprocessing step includes: After normalizing the operation data set, process the normalized operation data set and the fault label to obtain the fault label, and divide the normalized operation data set into an operation data training set and an operation data test set.

3. The fault diagnosis method according to claim 2, wherein The preprocessing step further includes: Initialize the neural network weights of the BP neural network with a Xavier uniform distribution to obtain the first-layer weight and the second-layer weight of the BP neural network; Form an activation function according to the normalized operation data set, the fault label, the initialized first-layer weight and the second-layer weight through the SoftSign function; Construct an error function according to the activation function, and calculate the total error value of the operation data training set through the error function.

4. The fault diagnosis method according to claim 2, wherein, The BP neural network optimization step includes: Calculate the additional momentum factor and the learning rate of the BP neural network according to the total error value; Train the BP neural network according to the learning rate of the BP neural network and the additional momentum factor to obtain the change amount of the first-layer weight and the change amount of the second-layer weight of the optimized BP neural network.

5. The fault diagnosis method according to claim 4, characterized in that, The BP neural network optimization step further includes: Calculate the first-layer weight and the second-layer weight of the optimized BP neural network through the backpropagation method according to the change amount of the first-layer weight and the change amount of the second-layer weight; Calculate the preliminary fault diagnosis result according to the first-layer weight and the second-layer weight.

6. The fault diagnosis method according to claim 2, wherein The fault type and fault probability prediction result obtaining step further includes: According to the preliminary fault diagnosis result, obtain the distance from the preliminary fault diagnosis result to the fault label through the distance measurement method of Euclidean distance; According to the nearest KNN&P algorithm with K = 1, take the fault label value that is closest to the preliminary result of the fault diagnosis.

7. A fault diagnosis system, characterized in that, The fault diagnosis system includes: Data acquisition unit: Collect the operation data of the tram supercapacitor thermal management system through sensors, construct an operation data set based on the operation data, and store the operation data in the cloud server through the Internet of Things platform; Data preprocessing unit: In the data training and analysis platform, process the operation data set to obtain the total error value of the operation data set; BP neural network optimization unit: Calculate the learning rate of the BP neural network according to the total error value, optimize the BP neural network according to the learning rate of the BP neural network and the additional momentum factor, and calculate the preliminary result of the fault diagnosis output by the optimized BP neural network according to the first-layer weight and the second-layer weight of the optimized BP neural network; Fault type and fault probability prediction result acquisition unit: Output the fault type and fault probability prediction result through the KNN&P fault diagnosis and probability prediction model according to the preliminary result of the fault diagnosis, and synchronize the fault type and the fault probability prediction result to the after-sales service platform; specifically, construct the KNN&P fault diagnosis and probability prediction model based on the nearest KNN&P algorithm with K = 1, test the KNN&P fault diagnosis and probability prediction model through the operation data test set, and obtain the fault type and the fault probability prediction result through the KNN&P fault diagnosis and probability prediction model after the sample set passes the test.

8. The fault diagnosis system according to claim 7, wherein The operation data set includes the operation data of the pipeline part of the thermal management system, the operation data of the electrical part of the thermal management system, the data of the third supercapacitor module, the operation state data of the devices of the thermal management system, and the working mode feedback data of the devices of the thermal management system. The data acquisition unit collects the operation state data of the devices of the thermal management system and the working mode feedback data through the IO of the thermal management system controller.

9. The fault diagnosis system according to claim 8, characterized in that The data acquisition unit further includes: Temperature sensor and pressure sensor, collect the operation data of the pipeline part of the thermal management system through the temperature sensor and the pressure sensor; Current sensor and voltage sensor, collect the operation data of the electrical part of the thermal management system through the current sensor and the voltage sensor; Humidity sensor and the temperature sensor, collect the data of the first supercapacitor module through the humidity sensor and the temperature sensor; Among them, the data acquisition unit obtains the data of the second supercapacitor module through the vehicle CAN communication, and analyzes the data of the first supercapacitor module and the data of the second supercapacitor module to obtain the third supercapacitor module.

Images