A test point design method for a waterjet propulsion device control system

Abstract

The application discloses a kind of water jet propulsion device control system test point design method, water jet propulsion device control system is formed by the concatenation of multiple functional circuit, for the system circuit of water jet propulsion device control system, design is by component level to circuit level Gradual test point arrangement method.The application uses the pulse signal of rich frequency component to stimulate system circuit under offline condition, and considering the tolerance problem of electrical components, multiple repeated sampling is carried out using Monte Carlo method, more fault characteristic information corresponding to the test point is retained, and the influence of poor robustness caused by noise in the design process is reduced;Based on clustering, an integer coding table is constructed, which no longer depends on the circuit transfer function and the fault threshold set by human, solves the problem of fuzzy group division, and is also suitable for large and complex circuits.

Description

Technical Field

[0001] This invention relates to a test point design method for a water jet propulsion device control system, and more particularly to a test point design method for fault diagnosis of analog circuits in a water jet propulsion device control system, belonging to the field of analog circuit fault diagnosis technology. Background Technology

[0002] Waterjet propulsion, as a novel power propulsion method, has been widely applied in various naval propulsion systems. A typical waterjet propulsion system mainly consists of three parts: a mechanical system, a control system, and a hydraulic system, with the control system being an electrical unit. Due to factors such as usage intensity, frequency, and operating environment, a series of performance degradation and malfunctions occur in the control system caused by component fatigue, such as component damage, decreased electrical performance, and circuit faults. This leads to deviations between the waterjet propulsion system's circuit angle and actual position, malfunctions, and incomplete control. While most malfunctions can be resolved by switching to manual emergency control mode, restarting the control system, and sending technicians to replace relevant computer boards, some malfunctions are indirect, intermittent, and cannot be reproduced, making it difficult to analyze and locate the true cause. This poses a risk to the ship's navigation safety and reduces its seaworthiness, making it imperative to improve the fault diagnosis capabilities of the waterjet propulsion system control system.

[0003] Test point design is a crucial step in fault diagnosis. Using only output nodes as test points may result in low test accuracy and fail to meet normal requirements. Given the large scale and numerous nodes in the waterjet propulsion system control system, using all nodes as test points would increase testing costs and time. As faults propagate through the system, the testing cost and fault identification capability of different test points vary. Therefore, a reasonable test point set design is essential to ensure that the system's operating state can be best distinguished with the fewest possible test points. Summary of the Invention

[0004] The technical problem to be solved by this invention is: how to improve the accuracy of testing and ensure that the system working status can be distinguished with the fewest number of test points.

[0005] To solve the above-mentioned technical problems, the technical solution of the present invention is to provide a test point design method for a water jet propulsion device control system. The method is characterized in that the water jet propulsion device control system is composed of multiple cascaded functional circuits. For the system circuit of the water jet propulsion device control system, a step-by-step test point arrangement method from the component level to the circuit level is designed. The test point design method for the water jet propulsion device control system includes the following steps:

[0006] Step 1: Build a simulation model of the water jet propulsion device control system and inject fault modes; Step 2: Select all nodes that can be brought out from the system circuit of the water jet propulsion device control system as test points, and obtain multiple sets of corresponding pulse response signals through Monte Carlo analysis; Step 3: Perform multi-level wavelet packet decomposition on all signals respectively, reconstruct the wavelet packet coefficients in different frequency bands, calculate the energy of each wavelet packet coefficient in the last level, and form an energy vector; Step 4: Construct a component-level fault dictionary encoding table for the system circuit based on the improved fault dictionary method; Step 5: Select the optimal set of test points; based on the criterion of isolating all fault types, select test points from the fault code table to form the optimal fault dictionary; Step 6: Arrange circuit-level test points. In each functional circuit, based on the component-level coded fault dictionary table, filter the circuits to determine whether a fault has occurred using the fewest test points. In the component-level coded fault table, when all fault row vectors are not equal to the normal state row vectors, it is considered that a fault has occurred.

[0007] Preferably, in step 3, the energy vector obtained by wavelet packet energy transformation contains the characteristics of the signal in the time domain, frequency domain, and time frequency domain, and can more comprehensively describe the characteristics of the signal.

[0008] Preferably, in step 1, the simulation model of the water jet propulsion device control system is built using general-purpose circuit analysis software.

[0009] Preferably, in step 3, the wavelet packet decomposition of the signal is performed using the following formula: , in Representative scale indicators. Represents frequency indicators. Represents variables, Represents location indicators, and Represents the coefficients of the multi-resolution filter. Indicates signal In a certain subspace wavelet packet coefficients on and They represent Two subspaces and Wavelet packet coefficients on; The reconstruction algorithm for wavelet packet coefficients is shown in the following equation; ; Z is the set of integers; The low-pass filter coefficients corresponding to index i-2l; The high-pass filter coefficients at index i-2l; and These represent the wavelet packet coefficients in the two subspaces of the next scale index.

[0010] Preferably, in step 3, the original signal is subjected to... Layer wavelet packet transform can obtain The wavelet packet coefficients; let the i-th wavelet packet coefficients be... The wavelet packet coefficient vector of the layer is After reconstructing all wavelet packet coefficients, the reconstructed signal vector is obtained. Reconstructed signal energy As shown in the following formula, , in, It is the wavelet packet transform layer. Indicates the index of the wavelet packet coefficients The number of signal sampling points is , Indicates the first The number of points in each wavelet packet coefficient in the layer. , Indicates the first The first reconstructed signal The amplitude of the point; This represents the total number of wavelet packet coefficients obtained after m-level wavelet packet decomposition.

[0011] Preferably, in step 3, the following is determined: The energy vector of the reconstructed signal is obtained from wavelet packets. After normalization, the wavelet packet energy feature vector is obtained. .

[0012] Preferably, the selection of the optimal test point set in step 5 includes the following steps: Step 5-1-1: Set the optimal set of measurement points Empty, alternative measurement point set Includes all measurement points in the fault dictionary; Step 5-1-2, from Take out Largest measuring point Put in Among them For measuring points The total number of faults that can be isolated, i.e., the number of fuzzy groups; Step 5-1-3, Check the addition After Can all faults be isolated? If so, then terminate. Step 5-1-4, All faults that can be isolated at all measurement points are removed from the fault dictionary; Step 5-1-5, Calculation All measuring points Value, here Value by and Determined together, and return to step 5-1-2; Step 5-2-1: Evaluate the effectiveness of test point set selection using the CH index of clustering. The CH index is defined based on the ratio of within-class to between-class variance. Internal clustering effect of classes It is the within-class discreteness matrix. It is the inter-class scatter matrix. For the sample size, Number of categories;

[0013] Step 5-2-2: The average CH value of the test point set is used as a metric. A higher value indicates better clustering density and separation, and a better fuzzy group partitioning effect. b The number of actual test points in the set; the formula for calculating the average CH value of the test point set is as follows: ; Step 5-2-3: Calculate the average CH value of the optimal test point set and the average CH value of all test point sets respectively. If the value of the optimal test point set is larger, it indicates that the test point design result is reasonable. Let be the CH index value corresponding to the j-th test point in the test point set.

[0014] Preferably, in step 4, in order to obtain more fault information, a pulse signal with rich frequency components is used to excite the circuit under test in the offline condition; since the pulse response amplitude components of the circuit are complex and not conducive to determining the fixed separable threshold of the fault, an improved algorithm of fault dictionary method is proposed, which uses feature clustering results to guide fuzzy group division instead of threshold setting.

[0015] Preferably, in step 4, the step of constructing the component-level fault dictionary encoding table of the system circuit based on the improved fault dictionary method is as follows: Step 4-1: For all test points, first list all energy vectors for multiple fault modes; Step 4-2: To avoid the problem of an overly complex fault dictionary caused by assigning independent categories to each fault mode, some fault modes with no significant differences in features are grouped into the same fuzzy group. Faults within the same fuzzy group are indistinguishable at the test point. The BIRCH clustering algorithm achieves an efficient clustering process through the hierarchical structure of the CF tree and the compression and splitting operations of clustering features. BIRCH hierarchical clustering is performed separately on the energy feature set of each test point. Step 4-3: Organize the fuzzy group division of all test points and obtain the fault code table for the system circuit.

[0016] Preferably, step 4-2 includes the following steps: Step 4-2-1: First, select a sample vector to generate the first CF, then add vectors one by one and update the structure of the CF tree; the CF tree is a multi-way tree, and each node represents a CF, which contains feature vector information and corresponding clustering statistics. Step 4-2-2: In order to reduce the size of the CF tree and improve the clustering efficiency, when adding a new CF to the CF tree, check whether there are similar CFs. If they exist, they are merged into a new CF. The similarity between CFs is achieved by calculating the Euclidean distance between feature vectors and setting a threshold. Step 4-2-3: When the number of nodes in the CF tree reaches a threshold or nodes can no longer be merged, perform cluster splitting. Divide the nodes of the CF tree, generate new child nodes, and select appropriate data points to be placed into the child nodes according to certain criteria; Step 4-2-4: Traverse the leaf nodes of the CF tree and generate the clusters as the final clustering result; Step 4-2-5: Define the cluster type with the largest number of clusters in each fault mode as the fuzzy group category, and number the fuzzy groups.

[0017] This invention proposes a test point design method for a waterjet propulsion device control system, arranging system test points step-by-step from the component level to the circuit level. Currently common methods based on integer encoding tables suffer from susceptibility to noise, difficulty in applying to large and complex circuits, and the need for manual parameter setting, resulting in poor robustness and generalization of the determined test point design, making them unsuitable for widespread use. This invention uses pulse signals with rich frequency components to excite the system circuit offline. Considering the tolerance issues of electrical components, it employs Monte Carlo methods for multiple repeated sampling, preserving more fault characteristic information corresponding to the test points and reducing the impact of noise on the design process's poor robustness. Based on clustering to construct an integer encoding table, it no longer relies on circuit transfer functions and manually set fault thresholds, solving the problem of difficult fuzzy group partitioning, and is also applicable to large and complex circuits.

[0018] Compared with the prior art, the present invention has the following advantages: (1) Use pulse signals to excite the circuit under test to obtain response signals with rich and continuously changing amplitude components to construct a fault dictionary and retain more fault feature information; (2) The clustering results are used to construct fuzzy groups of impulse response signals, while reducing the impact of manually setting voltage thresholds on the fuzzy group division effect; (3) Complete the test point layout from the component level to the circuit level, realize the differentiation of all fault modes with the fewest test points, and help improve the performance of the control system in subsequent fault monitoring and fault diagnosis tasks. Attached Figure Description

[0019] Figure 1 A schematic diagram of the basic components of a waterjet propulsion system control system; Figure 2 The circuit diagram of the signal conditioning unit and the available test points are shown. Figure 3 A flowchart for designing test points for the water jet propulsion device control system. Detailed Implementation

[0020] To make the present invention more apparent and understandable, preferred embodiments are described in detail below with reference to the accompanying drawings.

[0021] Common methods for designing test points include integer-encoded fault tables (fault dictionary methods), sensitivity matrix analysis, and optimization methods based on intelligent algorithms. However, current test point design methods suffer from drawbacks such as requiring precise circuit topology and the need for manual setting of some parameters or limiting factors. Consequently, the determined optimal test point set exhibits poor robustness and is not suitable for widespread use.

[0022] Therefore, this invention proposes a test point design method for a waterjet propulsion device control system. Based on an improved fault dictionary method using unsupervised clustering, it uses pulse signals with rich frequency components to excite the system circuit in offline conditions. This solves the problem of difficulty in dividing fuzzy groups using fault dictionary technology while obtaining more fault feature information. The fault dictionary method is a mature test point layout method. Improvements are made to this method by listing the pulse response features corresponding to all possible fault modes under all candidate test points. Fault modes with no significant differences in fault features under the same test point are grouped into the same fuzzy group. The fuzzy group division results under all test points are compiled into an integer-coded fault table, and the optimal test point set is selected according to a specific discrimination criterion. In the fuzzy group division stage, the balanced iterative reduction and clustering using hierarchical methods (BIRCH) are used to perform hierarchical clustering on the feature samples under each test point. The sample features are stored in the clustering feature (CF) of the nodes, which is used to form a clustering feature tree (CF Tree) to obtain the clustering results. Here, CF represents a triple, denoted by (N, LS, SS), where N represents the number of leaf nodes under this node, LS represents the sum vector of the features of each dimension of the leaf node samples under this node, and SS represents the sum of the squares of the features of each dimension of the leaf nodes contained under this node. Based on the BIRCH algorithm, large-scale datasets can be processed quickly to obtain the clustering results of all samples under each fault mode. The group with the largest number of belonging categories is the fuzzy group corresponding to that fault mode.

[0023] This invention proposes a test point design method for a water jet propulsion device control system, wherein the water jet propulsion device control system is composed of multiple cascaded functional circuits.

[0024] This paper studies the design of test points for the control system circuit of a water jet propulsion device. A step-by-step test point arrangement method from the component level to the circuit level is designed. The component-level test point design employs an improved integer encoding table test point design method based on impulse response and clustering (a fault dictionary test point design method based on BIRCH is used during the component-level test point design process). During the circuit-level test point design process, the selected test points from the component-level set are further screened, with the criterion being that the selected test point set can distinguish whether each circuit has experienced a fault.

[0025] An integer encoding table is constructed based on the response signal with rich and continuously changing amplitude components obtained by exciting the circuit under test with a pulse signal. Taking into account the tolerance of electrical components, multiple Monte Carlo samplings are performed to retain more fault characteristic information and improve the robustness of the test point design results.

[0026] The fuzzy groups are divided by clustering. Taking the BIRCH algorithm as an example, the fault modes of a measurement point are divided into fuzzy groups based on the BIRCH clustering results of the impulse response. If Monte Carlo response signals of the same fault mode belong to different fuzzy groups, the fuzzy group with the most numbers is used.

[0027] The present invention provides a method for designing measuring points in a waterjet propulsion device control system, which specifically includes the following steps: (1) A simulation model of the water jet propulsion device control system was built using PSPICE (General Circuit Analysis Program) software, and fault modes were injected; (2) Take all the nodes that can be brought out in the system circuit as test points and obtain multiple sets of corresponding impulse response signals through Monte Carlo analysis; (3) Perform multi-level wavelet packet decomposition on all signals to obtain wavelet packet coefficients in different frequency bands (i.e., reconstruct the wavelet packet coefficients), calculate the energy of each wavelet packet coefficient in the last level, and form a feature vector. The energy vector obtained after wavelet packet energy transformation contains the characteristics of the signal in the time domain, frequency domain, and time frequency domain, and can more comprehensively describe the characteristics of the signal; (3-1) First, perform wavelet packet decomposition on the signal, as shown in the following formula. Wherein Representative scale indicators. Represents frequency indicators. Represents variables, Represents location indicators, and Represents the coefficients of the multi-resolution filter. Indicates signal In a certain subspace wavelet packet coefficients on and They represent Two subspaces and Wavelet packet coefficients on;

[0028] (3-2) The reconstruction algorithm for wavelet packet coefficients is shown in the following equation;

[0029] (3-3) Suppose that the original signal is processed... Layer wavelet packet transform can obtain The wavelet packet coefficients; let the i-th wavelet packet coefficients be... The wavelet packet coefficient vector of the layer is After reconstructing all wavelet packet coefficients, the reconstructed signal vector is obtained. Reconstructed signal energy As shown in the following formula. Where, It is the wavelet packet transform layer. Indicates the index of the wavelet packet coefficients The number of signal sampling points is , Indicates the first The number of points in each wavelet packet coefficient in the layer. , Indicates the first The first reconstructed signal The amplitude of the point.

[0030]

[0031] (3-4) Find The energy vector of the reconstructed signal is obtained from wavelet packets. After normalization, the wavelet packet energy feature vector is obtained. .

[0032] (4) Construct a component-level fault dictionary encoding table for the system circuit based on the improved fault dictionary method. To obtain more fault information, the system circuit under test is excited by a pulse signal with rich frequency components in the offline condition. Since the pulse response amplitude components of the circuit are complex and not conducive to determining the fixed separable threshold of the fault, an improved fault dictionary method algorithm is proposed, which uses the feature clustering results to guide the fuzzy group division instead of the threshold setting; (4-1) For all test points, first list all energy vectors for various fault modes; (4-2) To avoid the problem of an overly complex fault dictionary caused by assigning independent categories to each fault mode, some fault modes with no significant differences in features are grouped into the same fuzzy group. Faults within the same fuzzy group are indistinguishable at the test point. The BIRCH clustering algorithm achieves an efficient clustering process through the hierarchical structure of the CF tree and the compression and splitting operations of clustering features, performing BIRCH hierarchical clustering on the energy feature set of each test point separately; (4-2-1) First, select a sample vector to generate the first CF (Clustered Flow), then add vectors one by one and update the structure of the CF tree. The CF tree is a multi-way tree, where each node represents a CF, containing feature vector information and corresponding clustering statistics. (4-2-2) In order to reduce the size of the CF tree and improve the clustering efficiency, when adding a new CF to the CF tree, it is checked whether there are similar CFs. If they exist, they are merged into a new CF. The similarity between CFs is achieved by calculating the Euclidean distance between feature vectors and setting a threshold. (4-2-3) When the number of nodes in the CF tree reaches a threshold or nodes can no longer be merged, cluster splitting is performed. The nodes of the CF tree are divided to generate new child nodes, and appropriate data points are selected and placed into the child nodes according to certain criteria; (4-2-4) Traverse the leaf nodes of the CF tree and generate the clusters as the final clustering result.

[0033] (4-2-5) It is stipulated that the cluster type with the largest number of clusters in each failure mode is the fuzzy group category, and the fuzzy groups are numbered; (4-3) Organize the fuzzy group division of all test points and obtain the fault code table under the system circuit; (5) Select the optimal set of test points. Based on the criterion of isolating all fault types, select test points from the fault coding table to form the optimal fault dictionary; (5-1-1) Let the optimal set of measurement points be set. Empty, alternative measurement point set Includes all measurement points in the fault dictionary; (5-1-2) From Take out Largest measuring point Put in Among them For measuring points The total number of faults that can be isolated, i.e., the number of fuzzy groups; (5-1-3) Check the addition After Can all faults be isolated? If so, then terminate. (5-1-4) will All faults that can be isolated at all measurement points are removed from the fault dictionary; (5-1-5) Calculation All measuring points Value, here Value by and Determined together, and returned (5-1-2); (5-2-1) The effectiveness of test point set selection is evaluated using the internal clustering metric Calinski-Harabasz (CH metric). The CH metric is defined based on the ratio of within-class to between-class variances. Internal clustering effect of classes It is the within-class discreteness matrix. It is the inter-class scatter matrix. For the sample size, For the number of categories, This refers to the fault category number;

[0034] (5-2-2) The average CH value of the test point set is used as a metric. The larger the value, the better the clustering density and separation, and the better the fuzzy group partitioning effect. b This represents the actual number of test points in the set. (Here, 's' represents the "average CH value of the test point set"). (5-2-3) Calculate the average CH value of the optimal test point set and the total test point set respectively. If the value of the optimal test point set is larger, it indicates that the test point design result is reasonable. (6) Arrange test points at the circuit level of the control system. In each functional circuit, based on the component-level coded fault dictionary table, the selection is carried out with the criterion of distinguishing whether the current circuit has a fault using the fewest test points. In the coded fault table, when all fault row vectors are not equal to the row vectors of the normal state, it is considered that whether a fault has occurred is distinguishable.

[0035] Example Figure 1 This paper describes the basic components of a waterjet propulsion system control system, which consists of sensors, a signal conditioning unit, a main control unit, a power drive unit, a CAN bus, actuators, and a host computer. The test point layout for the waterjet propulsion system control system is mainly carried out from the following three aspects: test points are placed at the outputs of the steering sensor, yaw angle sensor, and speed sensor; test points are set at the bus controller and transceiver of the CAN bus system, and the CAN bus operating status is monitored in real time using a bus comprehensive analyzer; for the cascaded circuit system composed of the signal conditioning unit, main control unit, and power drive unit, circuit test points are laid out using an improved fault dictionary method.

[0036] The embodiments of the present invention will be described in conjunction with the signal conditioning unit of the water jet propulsion device control system. The circuit diagram of the signal conditioning unit and the selectable test points are as follows: Figure 2 As shown, the specific steps are as follows: (1-1) A simulation model of the signal conditioning unit of the water jet propulsion device control system was built using PSPICE software, using an amplitude of 5V and a frequency of 2kHz. Using a pulse signal as an excitation source, sampling can obtain the response signal at the test point of the circuit to be selected; (1-2) Set the tolerance ranges of resistors and capacitors in the circuit to 5% and 10%, respectively. Through mechanism analysis, the common fault modes of the signal conditioning unit are signal drift and insufficient signal-to-noise ratio. Fault simulation can be performed by injecting faults into the passive components of the circuit. After sensitivity analysis, the fault locations were selected as C1, C2, R1, R3, and R4. The fault categories, nominal values, and fault values ​​of each component in the experimental circuit are shown in the table below:

[0037] (2) Through Monte Carlo analysis, 300 sets of impulse response signals were sampled for each fault category at each candidate test point, that is, the data dimension of the signal is 1000; (3) Using the Coiflet series coif1 discrete wavelet function as the wavelet basis, perform 5-level wavelet packet decomposition on each 1000-dimensional row vector to obtain wavelet packet coefficients in different frequency bands, calculate the energy of each wavelet packet coefficient in the 5th level, and form a 32-dimensional energy vector. After normalization, the wavelet packet energy feature vector is obtained. ; (4) Based on the improved fault dictionary method, a component-level fault dictionary encoding table for the system circuit is constructed. Since the pulse response amplitude components of the circuit are complex, it is not conducive to determining the fixed separable threshold of the fault. The feature clustering results are used to guide the fuzzy group division. (4-1) For all test points, first list all energy vectors for various fault modes; (4-2) First, for the energy feature vector of a certain test point, certain fault modes that are clustered together by BIRCH are classified into the same fuzzy group. Faults within the same fuzzy group are indistinguishable at the test point. The cluster type with the largest number of clusters is defined as the grouping category of a certain fault mode, thus obtaining the fuzzy group classification of fault modes at a certain test point; (4-3) Organize the fuzzy group division of all test points and obtain the fault code table for this circuit, as shown in the table below:

[0038] (5) Selecting the optimal test point set. The standard is to isolate all fault types with the fewest possible test points. In the fault coding table, when all fault row vectors are not equal to the normal state row vectors, it is considered that the fault is distinguishable and isolable. The optimal test point set is not unique; priority is given to selecting test points from those with larger clustering results. The optimal fault coding table is constructed by selecting test points {Vout1, Vout3, Vout7, Vout8} from the fault coding table, as shown in the table below:

[0039] (5-2-1) The effectiveness of test point set selection is evaluated using the internal clustering metric Calinski-Harabasz (CH metric). The CH metric is defined based on the ratio of within-class to between-class variances. Internal clustering effect of classes It is the within-class discreteness matrix. It is the inter-class scatter matrix. For the sample size, This represents the number of categories. The CH index for each test point is shown in the table below:

[0040]

[0041] (5-2-2) The average CH value of the test point set is used as a measure. The larger the value, the better the clustering density and separation, and the better the fuzzy group partitioning effect. Where b is the actual number of test points in the set.

[0042] (5-2-3) Calculate the average CH value of the optimal test point set and the set of all test points respectively. The set with the larger value is more reasonable. In this embodiment, the average CH values ​​of the optimal test point set and the set of all test points are 6033 and 4121.3 respectively, and the test point design result is reasonable. (6) When arranging the test points for the cascaded circuit of the signal conditioning unit, main control unit and power drive unit, the selection is based on the component-level coding fault dictionary table of the three circuits, with the standard of using the fewest test points to distinguish whether the current circuit has a fault. Similarly, the test points with larger clustering results labels are selected first. In the coding fault table, when all fault row vectors are not equal to the row vectors of the normal state, it is considered that whether a fault has occurred is distinguishable.

Claims

1. A method for designing test points for a waterjet propulsion device control system, characterized in that, The water jet propulsion device control system is composed of multiple cascaded functional circuits. For the system circuit of the water jet propulsion device control system, a step-by-step test point arrangement method from the component level to the circuit level is designed. The test point design method for the water jet propulsion device control system includes the following steps: Step 1: Build a simulation model of the water jet propulsion device control system and inject fault modes; Step 2: Select all nodes that can be brought out from the system circuit of the water jet propulsion device control system as test points, and obtain multiple sets of corresponding pulse response signals through Monte Carlo analysis; Step 3: Perform multi-level wavelet packet decomposition on all signals respectively, reconstruct the wavelet packet coefficients in different frequency bands, calculate the energy of each wavelet packet coefficient in the last level, and form an energy vector; Step 4: Construct a component-level fault dictionary encoding table for the system circuit based on the improved fault dictionary method; Step 5: Select the optimal set of test points; based on the criterion of isolating all fault types, select test points from the fault code table to form the optimal fault dictionary; Step 6: Arrange circuit-level test points. In each functional circuit, based on the component-level coded fault dictionary table, filter the circuits to determine whether a fault has occurred using the fewest test points. In the component-level coded fault table, when all fault row vectors are not equal to the normal state row vectors, it is considered that a fault has occurred. In step 4, the steps for constructing the component-level fault dictionary encoding table of the system circuit based on the improved fault dictionary method are as follows: Step 4-1: For all test points, first list all energy vectors for each of the various fault modes; Step 4-2: To avoid the problem of an overly complex fault dictionary caused by assigning independent categories to each fault mode, some fault modes with no significant differences in features are grouped into the same fuzzy group. Faults within the same fuzzy group are indistinguishable at the test point. The BIRCH clustering algorithm achieves an efficient clustering process through the hierarchical structure of the CF tree and the compression and splitting operations of clustering features. BIRCH hierarchical clustering is performed separately on the energy feature set of each test point. Step 4-3: Organize the fuzzy group division of all test points and obtain the fault code table for the system circuit.

2. The test point design method for a waterjet propulsion device control system as described in claim 1, characterized in that, In step 3, the energy vector obtained by wavelet packet energy transformation contains the characteristics of the signal in the time domain, frequency domain, and time frequency domain, and can more comprehensively describe the characteristics of the signal.

3. The test point design method for a waterjet propulsion device control system as described in claim 1, characterized in that, In step 1, the simulation model of the water jet propulsion device control system is built using general-purpose circuit analysis software.

4. The test point design method for a waterjet propulsion device control system as described in claim 1, characterized in that, In step 3, the wavelet packet decomposition of the signal is performed using the following formula. ， in Representative scale indicators. Represents frequency indicators. Represents variables, Represents location indicators, and Represents the coefficients of the multi-resolution filter. Indicates signal In a certain subspace wavelet packet coefficients on and They represent Two subspaces and Wavelet packet coefficients on; The reconstruction algorithm for wavelet packet coefficients is shown in the following equation; ； Z is the set of integers; The low-pass filter coefficients corresponding to index i-2l; The high-pass filter coefficients at index i-2l; and These represent the wavelet packet coefficients in the two subspaces of the next scale index.

5. The test point design method for a waterjet propulsion device control system as described in claim 4, characterized in that, In step 3, it is assumed that the original signal is processed... Layer wavelet packet transform can obtain The wavelet packet coefficients; let the i-th wavelet packet coefficients be... The wavelet packet coefficient vector of the layer is After reconstructing all wavelet packet coefficients, the reconstructed signal vector is obtained. Reconstructed signal energy As shown in the following formula, ， in, Number of wavelet packet transform layers Indicates the index of the wavelet packet coefficients The number of signal sampling points is , Indicates the first The number of points in each wavelet packet coefficient in the layer. , Indicates the first The first reconstructed signal The amplitude of the point; This represents the total number of wavelet packet coefficients obtained after m layers of wavelet packet decomposition.

6. The test point design method for a waterjet propulsion device control system as described in claim 5, characterized in that, In step 3, the following is calculated: The energy vector of the reconstructed signal is obtained from wavelet packets. After normalization, the wavelet packet energy feature vector is obtained. .

7. The test point design method for a waterjet propulsion device control system as described in claim 1, characterized in that, Step 5, selecting the optimal test point set, includes the following steps: Step 5-1-1: Set the optimal set of measurement points Empty, alternative measurement point set Includes all measurement points in the fault dictionary; Step 5-1-2, from Take out Largest measuring point Put in Among them For measuring points The total number of faults that can be isolated, i.e., the number of fuzzy groups; Step 5-1-3, Check the addition After Can all faults be isolated? If so, then terminate. Step 5-1-4, All faults that can be isolated at all measurement points are removed from the fault dictionary; Step 5-1-5, Calculation All measuring points Value, here Value by and Determined together, and return to step 5-1-2; Step 5-2-1: Evaluate the effectiveness of test point set selection using the CH index of clustering. The CH index is defined based on the ratio of within-class to between-class variance. Internal clustering effect of classes It is the within-class discreteness matrix. It is the inter-class scatter matrix. For the sample size, Number of categories; Step 5-2-2: The average CH value of the test point set is used as a metric. A higher value indicates better clustering density and separation, and a better fuzzy group partitioning effect. b The number of actual test points in the set; the formula for calculating the average CH value of the test point set is as follows: ； Step 5-2-3: Calculate the average CH value of the optimal test point set and the average CH value of all test point sets respectively. If the value of the optimal test point set is larger, it indicates that the test point design result is reasonable. Let be the CH index value corresponding to the j-th test point in the test point set.

8. The test point design method for a waterjet propulsion device control system as described in claim 1, characterized in that, In step 4, to obtain more fault information, the circuit under test is excited by a pulse signal with rich frequency components in the offline condition. Since the pulse response amplitude components of the circuit are complex and not conducive to determining the fixed separable threshold of the fault, an improved algorithm of fault dictionary method is proposed, which uses feature clustering results to guide fuzzy group division instead of threshold setting.

9. The test point design method for a waterjet propulsion device control system as described in claim 1, characterized in that, Step 4-2 includes the following steps: Step 4-2-1: First, select a sample vector to generate the first CF, then add vectors one by one and update the structure of the CF tree; the CF tree is a multi-way tree, and each node represents a CF, which contains feature vector information and corresponding clustering statistics. Step 4-2-2: In order to reduce the size of the CF tree and improve the clustering efficiency, when adding a new CF to the CF tree, check whether there are similar CFs. If they exist, they are merged into a new CF. The similarity between CFs is achieved by calculating the Euclidean distance between feature vectors and setting a threshold. Step 4-2-3: When the number of nodes in the CF tree reaches the threshold or the nodes can no longer be merged, perform cluster splitting; divide the nodes of the CF tree, generate new child nodes, and select appropriate data points to put into the child nodes according to certain criteria. Step 4-2-4: Traverse the leaf nodes of the CF tree and generate the clusters as the final clustering result; Step 4-2-5: Define the cluster type with the largest number of clusters in each fault mode as the fuzzy group category, and number the fuzzy groups.

Images