Interactive Fault Diagnosis Method for Multi-Signal Flow Models

Abstract

The present invention relates to an interactive fault diagnosis method suitable for a multi-signal flow model, which generates a test-module adjacency matrix based on the multi-signal flow model: the adjacency matrix is ​​converted to obtain a reachable matrix between each LRU module; the equipment fault module-test correlation fault diagnosis matrix is ​​derived from the adjacency matrix of the LRU module by analyzing the correlation between the test point where each test is located and the LRU module signal and the reachability relationship between the LRU module and the test point; and the fault diagnosis matrix is ​​compressed and stored; at the beginning of the fault diagnosis, the serialized fault diagnosis matrix is ​​extracted from the stored database for deserialization and decompression, and then the fault diagnosis matrix is ​​simplified according to the existing fault symptoms of the current system and the available test tools on site, and a multi-step information heuristic algorithm based on fault probability information entropy and multi-weight priority is used to perform test recommendation, and interactive fault diagnosis is performed. A high-quality platform is provided for the practical use of fault diagnosis based on the multi-signal flow model.

Description

Technical Field

[0001] The invention relates to a ship control technology, and in particular to an interactive fault diagnosis method suitable for a multi-signal flow model. Background Art

[0002] The number of ships and the complexity of control systems are constantly increasing. As the main propulsion system of ships, adjustable pitch propellers are widely used. However, with the complexity of adjustable pitch propeller systems and the sharp increase in the number of applications, the disadvantages of high maintenance costs and difficulty are gradually emerging. Once the control system fails, the ship will be in danger of running aground, stranding, and colliding. Especially in complex waters, the consequences of losing control are extremely serious. The fault diagnosis research on the adjustable pitch propeller pitch electro-hydraulic control system has become an urgent problem to be solved. However, there are relatively few research results on system-level fault diagnosis of electro-hydraulic coupling systems at home and abroad. The diagnosis methods based on neural networks, support vector machines, etc. cannot be applied to systems with few diagnostic history samples such as adjustable pitch propellers. The model-based multi-signal flow theory is a relatively mature test modeling theory. The use of this theory for fault diagnosis can enable the ship's adjustable pitch propeller to have rapid diagnosis capabilities in a short time. Therefore, it is very necessary and meaningful to study a set of interactive fault diagnosis methods suitable for multi-signal flow models, and the application prospects are very broad. Summary of the invention

[0003] Aiming at the problem of applying multi-signal flow theory to ship fault diagnosis, an interactive fault diagnosis method suitable for multi-signal flow model is proposed.

[0004] The technical solution of the present invention is: an interactive fault diagnosis method applicable to a multi-signal flow model, which specifically comprises the following steps:

[0005] 1) Generate test-module adjacency matrix based on multi-signal flow model:

[0006] A multi-signal flow model is established through historical data, and the adjacency matrix between each LRU module is obtained by traversing the directed connection relationship of the multi-signal flow model. The adjacency matrix is ​​the adjacency relationship between each LRU module and the test point of the system;

[0007] 2) Generate a reachability matrix: The adjacency matrix obtained in step 1) is transformed to obtain the reachability matrix between each LRU module, that is, the functional signal reachability between each module;

[0008] 3) Generate fault diagnosis matrix: Through the correlation between the test point where each test is located and the LRU module signal and the reachability relationship between the LRU module and the test point, the equipment fault module-test correlation fault diagnosis matrix is ​​derived from the adjacency matrix of the LRU module;

[0009] 4) After obtaining the fault diagnosis matrix, the fault diagnosis matrix is ​​serialized and persistently stored according to the test columns, and compressed and stored;

[0010] 5) Start diagnosis: At the beginning of fault diagnosis, the method extracts the serialized fault diagnosis matrix from the database stored in step 4) for deserialization and decompression, and then simplifies the fault diagnosis matrix according to the existing fault symptoms of the current system and the available test tools on site. A multi-step information heuristic algorithm based on fault probability information entropy and multi-weight priority is used for test recommendation. According to the measurement points and test methods recommended by the generated diagnostic strategy, interactive fault diagnosis is performed to finally determine the fault location.

[0011] Furthermore, the multi-signal flow model is a model that uses a hierarchical directed graph to represent the signal flow orientation and the composition and interconnection relationship of each component unit, and characterizes the correlation between system composition, function, fault and test by defining the correlation between signals and component units, tests and signals.

[0012] Further, the derivation process of step 3) is as follows:

[0013] 3.1) Use each LRU module as a row vector and each basic test as a column vector to construct a fault module-test correlation matrix. When initializing, set all matrix element values ​​to 0;

[0014] 3.2) The correlation matrix is ​​assigned as follows: first, obtain the signal list contained in each LRU module and the basic test in the matrix, then take the test of each column in turn, query the test point to which the test belongs, and obtain the reachability relationship between the test point and the module from the reachability matrix. If the test point and the module are reachable and the intersection of the signal list of the test and the module is not empty, then the element value of the correlation matrix between the test and the module is set to 1.

[0015] Furthermore, the equipment fault module-test correlation matrix represents the correlation between the test results of each system test and the LRU module. When the test result is normal, the LRU modules related to the test in the correlation matrix are all functioning normally, that is, the element value is 1; when the test result is abnormal, the functions of the LRU modules related to the test in the correlation matrix may be abnormal. Each time a test is performed, the matrix rows with possible abnormalities are retained according to the test results, and the matrix rows with normal functions are deleted, thereby continuously reducing the number of matrix rows until it cannot be reduced. At this time, the LRU modules of the remaining row vectors are possible faulty modules.

[0016] Furthermore, the multi-step information heuristic algorithm based on fault probability information entropy and multi-weight priority is adopted: the fault probabilities of all possible fault modules of the current system are normalized and brought into the information entropy as the weight of the module fault information, so as to select the test with the largest reduction in system information entropy for detection, and the test priority, test cost, test time, and information entropy are weighted and added to calculate the total recommended probability. The preferences for test priority, test cost, test time, and information entropy weight can be set and updated at any time during the diagnosis process. The test preferences include test cost priority, test time priority, test priority priority, and default priority strategy.

[0017] The beneficial effects of the present invention are as follows: the present invention is applicable to an interactive fault diagnosis method of a multi-signal flow model, improves the diagnostic efficiency based on multi-signal flow fault diagnosis, provides real-time selection capability of diagnostic preferences during the diagnostic process, and provides a high-quality platform for the actual use of fault diagnosis based on a multi-signal flow model. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 The interactive fault diagnosis flow chart of the present invention is applicable to a multi-signal flow model;

[0019] Figure 2 A flow chart is generated for the diagnostic strategy method in the method of the present invention. DETAILED DESCRIPTION

[0020] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0021] like Figure 1 The interactive fault diagnosis flowchart applicable to the multi-signal flow model shown in the figure includes the following steps:

[0022] 1. First, the test-module adjacency matrix is ​​generated based on the multi-signal flow model: the multi-signal flow model is a model that uses a hierarchical directed graph to represent the signal flow orientation and the composition and interconnection relationship of each component unit (failure mode), and defines the correlation between signals (functions) and components (failure modes), tests and signals to characterize the correlation between system composition, functions, failures and tests. The multi-signal flow model is established through historical data. By deeply traversing the directed connection relationship of the multi-signal flow model, the adjacency matrix between each LRU (Line Replaceable Unit) module can be obtained, as shown in the following formula:

[0023]

[0024] Where s is the total number of system LUR and test points, m i For all modules and test points of the system LRU (such as power supply, power output port test point, relay, etc.), G s×s is the adjacency matrix between the system LRU modules and test points, g ij The value is 0 or 1. 0 means there is no adjacency between modules or between modules and tests, and 1 means there is an adjacency.

[0025] 2. Generate a reachable matrix: By converting the adjacency matrix, we can get the reachable matrix between each LRU module, that is, the functional signal reachability between each module. The conversion formula and reachable matrix are as follows:

[0026] tmpG=G+G 2 +G 3 +...+G s (2)

[0027]

[0028]

[0029] The above expression is the conversion formula between the reachable matrix and the adjacency matrix. The intermediate matrix tmpG can be derived from formula (2). The reachable matrix rechG can be obtained by converting the element values ​​in the intermediate matrix tmpG into binarized values ​​using the piecewise function of formula (3).

[0030] 3. Generate fault diagnosis matrix: By analyzing the correlation between the test points and LRU module signals of each test and the reachability relationship between the LRU module and the test points, the equipment fault module-test correlation fault diagnosis matrix can be derived from the adjacency matrix of the LRU module. The specific derivation process is as follows:

[0031] 3.1. Using each LRU module as a row vector and each basic test as a column vector, construct the following fault module-test correlation matrix D: m×m , where m is the number of LRU modules in the system, l i is the system LRU module, n is the number of basic tests in the system, t i For system basic tests (such as voltage detection of power output port, power supply internal resistance detection, etc.), d ij To equip the LRU fault module - test correlation information, its value is 0 or 1, and the matrix element values ​​are all set to 0 during initialization.

[0032]

[0033] 3.2. The correlation matrix is ​​assigned as follows: first, obtain the signal list contained in each LRU module and the basic test in the matrix, then take the test of each column in turn, query the test point to which the test belongs, and obtain the reachability relationship between the test point and the module from the reachability matrix. If the test point and the module are reachable and the intersection of the signal list of the test and the module is not empty, then the element value of the correlation matrix between the test and the module is set to 1.

[0034] 4. After obtaining the fault diagnosis matrix, the fault diagnosis matrix is ​​serialized and persistently stored according to the test columns, and compressed and stored. The specific serialization and compression method is: convert the matrix binary value into hexadecimal system with 4 bits as a unit, and fill the last digit with 0 if it is less than 4 bits, and store it in the database in the form of a string.

[0035] The equipment fault module-test correlation matrix represents the correlation between the test results of each system test and the LRU module. When the test result is normal, the LRU modules related to the test (i.e., the element value is 1) in the correlation matrix are all functioning normally. When the test result is abnormal, the LRU modules related to the test in the correlation matrix are all likely to be abnormal. For the fault diagnosis of the adjustable pitch propeller, each time a test is performed, the matrix rows that may be abnormal will be retained according to the test results, and the matrix rows that function normally will be deleted, thereby continuously reducing the number of matrix rows until it cannot be reduced. At this time, the LRU modules of the remaining row vectors are possible fault modules, and fault diagnosis can be performed directly.

[0036] The process of fault diagnosis based on the fault diagnosis correlation matrix can be equivalent to the set partitioning problem. It can be proved that the construction problem of the optimization decision tree is an NP-complete problem. A heuristic search method based on information entropy can be used to obtain a better solution. The improvement of the present invention lies in the use of module failure probability as the basic probability parameter for information entropy calculation, the use of a multi-step heuristic method in the heuristic search to improve the method capability, and the addition of other influencing factors in combination with a multi-weight priority method to dynamically regulate the diagnosis tree generation process.

[0037] 5. Start diagnosis: At the beginning of fault diagnosis, the method extracts the serialized fault diagnosis matrix from the database stored in step 4, deserializes and decompresses it, and then simplifies the fault diagnosis matrix according to the existing fault symptoms of the current system and the available test tools on site, and then executes Figure 2 The diagnostic strategy method generation process shown in the figure adopts a multi-step information heuristic algorithm based on fault probability information entropy and multi-weight priority to make test recommendations. According to the measuring points and test methods recommended by the generated diagnostic strategy, interactive fault diagnosis is performed to finally determine the fault location.

[0038] The multi-step information heuristic diagnosis strategy generation method based on fault probability information entropy and multi-weight priority consists of three parts: a single-step information heuristic search method based on fault probability information entropy, a multi-step information heuristic search based on single-step information heuristic, and a multi-weight priority method. The specific methods and execution order of each part are as follows:

[0039] 1) Single-step information-inspired search method based on fault probability information entropy:

[0040] Shannon proposed that information is the elimination of uncertainty. The more uncertainty is eliminated, the greater the amount of information obtained. Information entropy can also be used as a measure of the complexity of a system. If the system is more complex and there are more different types of situations, then its information entropy is relatively large. In essence, the diagnosis of the controllable pitch propeller system is to simplify the system from the information obtained from each test result, thereby reducing the uncertainty of the faulty component. For the controllable pitch propeller system components, the probability that the module may fail in the system is related to the failure probability of the module itself. Therefore, this algorithm normalizes the failure probability of all possible faulty modules in the current system, and brings it into the information entropy formula as the weight of the module failure information, so as to select the test with the largest reduction in system information entropy for detection.

[0041] The single-step information-inspired search method based on the information entropy of the failure probability takes the maximum entropy reduction of the failure probability of a single test as the test selection criterion. In this method, the row elements (i.e., LRU modules) remaining in the current correlation matrix are called fuzzy sets x, and the test t k is chosen if this test maximizes the entropy reduction of the failure probability of a single test, then:

[0042] IG(x,t j )=-{p(x jp )log2p(x jp )+p(x jp )log2(x jf )} (6)

[0043] k=arg max j {IG(x,t j )} (7)

[0044] where x jp and x jf is the fuzzy set x in the test t j After the two result sets are divided, p(x jp ) and p(x jf ) is the sum of the normalized failure probabilities of the modules in the two result sets, IG(x,t j ) is the fuzzy group entropy reduction value of the test on the system, and k is the test number with the largest fuzzy group entropy reduction value on the system.

[0045] 2) Multi-step information heuristic search method based on single-step information heuristic:

[0046] The single-step information heuristic search method based on fault probability information entropy only determines the quality of the test results through a single-layer result, and its results are not good for predicting subsequent steps, and subsequent tests will affect the final diagnosis results. Therefore, the multi-step information heuristic search method based on single-step information heuristic can play a greater role in optimizing the test recommendation results. The steps of the multi-step information heuristic search method based on single-step information heuristic are as follows:

[0047] Step 1: Select a test from the available test set i , t i Split the current fault set x into two subsets according to the pass or fail of the output results.

[0048] Step 2: For each subset, select the best available test based on the above single-step information heuristic search, and split the subset according to the pass and fail output results.

[0049] Step 3: Recursively repeat step 2 for each subset until the depth of the partial diagnosis tree reaches a certain depth step or the fault set cannot be divided any further (no fuzzy group). Calculate the single-step average fault probability information entropy reduction of the partial diagnosis tree.

[0050] Step 4: Repeat steps 2 and 3 for each test selected in step 1 to obtain the single-step average failure probability information entropy reduction of all tests in the current test set.

[0051] Step 5: For the current fuzzy set, select the test with the largest reduction in single-step average failure probability information entropy as the optimal test.

[0052]

[0053]

[0054] 3) Multi-weight priority method:

[0055] In the process of fault diagnosis, the on-site environment is changeable, and the optimal step diagnosis tree obtained based on the module failure probability is not necessarily the optimal solution. Other influencing factors include test time, test cost, test expense and other factors (for example, the test of the high-mounted oil tank of the controllable pitch propeller is very difficult and costly, and it is not recommended to measure; when measuring the pressure at the valve port of the reversing valve, the system needs to be shut down to connect the pressure gauge, and the test cost is high). Therefore, the present invention further optimizes step 5 in the multi-step information heuristic search method, and changes the selection function to a multi-weight priority function. The function method is as follows:

[0056]

[0057] Among them, f1 is the weight value of the test information entropy reduction in the selection, f2 is the weight value of the test time cost in the selection, f3 is the weight value of the test cost in the selection, f4 is the weight value of the test subjective priority in the selection, t j (time) is the test t j The time cost required for execution, t j (cost) is the test t j Cost of execution, t j (Priority) is the test t j The subjective priority setting is finally selected to select the optimal test under the current failure mode set as the current recommended test execution Figure 1 The troubleshooting process of

[0058] The information entropy calculation of the algorithm is performed by substituting the normalized failure probability of each isolation module into the calculation, and the test priority, test cost, test time, and information entropy are weighted and added to calculate the total recommended probability. The preferences for test priority, test cost, test time, and information entropy weight can be set and updated at any time during the diagnosis process. The test preferences include test cost priority, test time priority, test priority priority, and default priority strategies.

[0059] The above-mentioned embodiments only express several implementation methods of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

Claims

1. An interactive fault diagnosis method suitable for a multi-signal flow model, characterized in that: The specific steps include: 1) Generate test-module adjacency matrix based on multi-signal flow model: A multi-signal flow model is established through historical data, and the adjacency matrix between each LRU module is obtained by traversing the directed connection relationship of the multi-signal flow model. The adjacency matrix is ​​the adjacency relationship between each LRU module and the test point of the system; 2) Generate a reachability matrix: The adjacency matrix obtained in step 1) is transformed to obtain the reachability matrix between each LRU module, that is, the functional signal reachability between each module; 3) Generate a fault diagnosis matrix: By analyzing the correlation between the test points at each test location and the LRU module signal and the reachability relationship between the LRU module and the test point, the equipment fault module-test correlation fault diagnosis matrix is ​​derived from the adjacency matrix of the LRU module; The derivation process is as follows: 3.1) Using each LRU module as a row vector and each basic test as a column vector, construct an equipment fault module-test correlation matrix. When initializing, all matrix element values ​​are set to 0; 3.2) The correlation matrix is ​​assigned as follows: first, the signal list of each LRU module and the basic test in the matrix is ​​obtained, and then the test of each column is taken in turn to query the test point to which the test belongs, and the reachability relationship between the test point and the module is obtained from the reachability matrix. If the test point and the module are reachable and the intersection of the signal list of the test and the module is not empty, the element value of the correlation matrix between the test and the module is set to 1; The equipment fault module-test correlation matrix represents the correlation relationship between each test result of the system and the LRU module. When the test result is normal, the LRU modules related to the test in the test correlation matrix are all functioning normally, that is, the element value is 1; when the test result is abnormal, the LRU modules related to the test in the test correlation matrix are all likely to be abnormal. Each time a test is performed, the matrix rows with possible abnormalities are retained according to the test result, and the matrix rows with normal functions are deleted, thereby continuously reducing the number of matrix rows until it cannot be reduced. At this time, the LRU modules of the remaining row vectors are possible faulty modules. 4) After obtaining the fault diagnosis matrix, the fault diagnosis matrix is ​​serialized and persistently stored according to the test columns, and compressed and stored; 5) Start diagnosis: At the beginning of fault diagnosis, the method extracts the serialized fault diagnosis matrix from the database stored in step 4) for deserialization and decompression, and then simplifies the fault diagnosis matrix according to the existing fault symptoms of the current system and the available test tools on site. A multi-step information heuristic algorithm based on fault probability information entropy and multi-weight priority is used for test recommendation. According to the measurement points and test methods recommended by the generated diagnostic strategy, interactive fault diagnosis is performed to finally determine the fault location.

2. The interactive fault diagnosis method applicable to a multi-signal flow model according to claim 1, characterized in that: The multi-signal flow model is a model that uses a layered directed graph to represent the signal flow orientation and the composition and interconnection relationship of each component unit, and characterizes the correlation between system composition, function, fault and test by defining the correlation between signals and component units, tests and signals.

3. The interactive fault diagnosis method applicable to a multi-signal flow model according to claim 1, characterized in that: The multi-step information heuristic algorithm based on fault probability information entropy and multi-weight priority is adopted: the fault probabilities of all possible fault modules of the current system are normalized and brought into the information entropy as the weight of the module fault information, so as to select the test with the largest reduction in system information entropy for detection, and the test priority, test cost, test time, and information entropy are weighted and added to calculate the total recommended probability. The preferences for test priority, test cost, test time, and information entropy weight can be set and updated at any time during the diagnosis process. The test preferences include test cost priority, test time priority, test priority priority and default priority strategy.

Images