CN110262466A - A kind of winged control fault detection and diagnosis method based on random forest - Google Patents

Abstract

The invention belongs to fly control fault diagnosis technology field, it is related to a kind of winged control fault detection and diagnosis method based on random forest.The present invention analyzes the composition of all fault trees first, error code corresponding to every fault tree and its logical relation, training sample is being formed using above-mentioned data, the training for completing Random Forest model, in flight control system fault diagnosis in the future, it is only necessary to which fault list is input in system, true fault can be navigated to, even ground crew can also be quickly completed the diagnostic operation to flight control system, and maximum time internal segment about Diagnostic Time improves working efficiency.The present invention is for diagnosing flight control system failure.

Description

A Flight Control Fault Detection and Diagnosis Method Based on Random Forest

technical field

The invention belongs to the technical field of flight control fault diagnosis, and relates to a random forest-based flight control fault detection and diagnosis method.

Background technique

The flight control system is the core system of the aircraft. The failure of its system components not only affects the performance of the flight control system, but also poses a great threat to the flight safety of the aircraft. Intelligent fault diagnosis technology is applied to the failure of the aircraft flight control system In the diagnosis, assisting the maintenance personnel to eliminate the faults of the flight control system in time, improving the maintenance efficiency of the aircraft, and ensuring the safe flight of the aircraft are currently urgently needed research contents.

Contents of the invention

The content to be solved by the present invention is to provide a method for detecting and diagnosing flight control faults based on random forest.

The technical problem to be solved in the present invention is:

The flight control fault detection and diagnosis method based on random forest includes the following steps:

Step 1. Analyze all real faults contained in the flight control system, the fault codes corresponding to each fault type and the relationship between the fault codes, and establish a fault tree;

Step 2. Take the known fault type as the data sample, use the fault code corresponding to the fault type as the characteristic variable of the data sample, and use the real fault corresponding to each fault type as the expected output to complete the training of multiple decision trees, A random forest is composed of multiple decision trees as described:

Step 3. Obtain the fault list that the current flight control system bursts out, and input the fault codes in the list into the random forest model, and each of the decision trees outputs the corresponding prediction results, and performs the prediction results of all the decision tree outputs Voting analysis, the random forest model finally outputs the real fault with the highest probability.

As a further improvement of the above technical solution, in the step 1, the following operation steps are made to the current fault state of the flight control system:

Step 101. Obtain all fault trees;

Step 102. Extract all the fault codes in the fault tree, and arrange and combine them according to the AND or non-logic between the faults to generate multiple fault samples;

Step 103. Standardize the fault code;

As a further improvement of the above technical solution, in the step 2, the following operation steps are made to the current fault state of the flight control system:

Step 201. Extract n data samples with the same sample capacity from all the data samples by the method of flush sampling with replacement, as the training samples for training the decision tree;

Step 202. Randomly extract feature variables of m from a training sample;

Step 203. At the internal nodes of the decision tree, select a feature Xi with the best classification effect from the m feature variables according to the principle of the minimum Gini impurity, and divide the node into two branches. The Gini impurity principle is Where P(i) represents the proportion of each category to the total number of categories;

Step 204. Repeat the operation described in step 203 for each internal node of the decision tree until the decision tree can accurately classify the training samples or the Gini impurity of each node of the decision tree reaches the minimum;

Step 205. Select the next training sample, repeat step 202 to step 204, until the decision tree corresponding to the purchased training sample is constructed;

Step 206. The decision trees constructed from the n training samples together form a random forest model, and the training of the random forest model is completed.

The beneficial effects of this statement are: the present invention first analyzes the composition of all fault trees, the fault codes corresponding to each fault tree and its logical relationship, and uses the above data to form training samples to complete the training of the random forest model. In the fault diagnosis of the flight control system, the real fault can be located only by inputting the fault list into the system. Even the ground crew can quickly complete the diagnosis operation of the flight control system, saving the diagnosis time in the maximum time and improving the work efficiency. efficiency. The invention is used for diagnosing the failure of the flight control system.

Description of drawings

Fig. 1 is a flow chart of the fault diagnosis scheme of the present invention.

Fig. 2 is a specific implementation flow chart of step 2 of the present invention.

Detailed ways

The method of the present invention will be described in further detail below in conjunction with the accompanying drawings.

With reference to Fig. 1 and Fig. 2, the present invention discloses a kind of flight control system fault diagnosis method of random algorithm, comprises the following steps:

Step 1. Analyze all the real faults contained in the flight control system, the fault codes corresponding to each fault type and the relationship between the fault codes;

Step 2. Take the known fault type as the data sample, use the fault code corresponding to the fault type as the characteristic variable of the data sample, and use the real fault corresponding to each fault type as the expected output to complete the training of multiple decision trees, A random forest is composed of multiple decision trees as described:

Step 3. Obtain the fault list that the current flight control system bursts out, and input the fault codes in the list into the random forest model, and each of the decision trees outputs the corresponding prediction results, and performs the prediction results of all the decision tree outputs Voting analysis, the random forest model finally outputs the real fault with the highest probability.

Specifically, the present invention first analyzes the composition of all fault trees, the fault codes corresponding to each fault tree and its logical relationship, and uses the above data to form training samples to complete the training of the random forest model. In the diagnosis, only the fault list needs to be input into the system, and the real fault can be located. Even the ground crew can quickly complete the diagnosis operation of the flight control system, saving the diagnosis time in the maximum time and improving the work efficiency.

Further as a preferred embodiment, in the concrete implementation of the present invention, described step 2 is to the following operation steps to the data processing of fault tree:

Step 101. Obtain all fault trees;

Step 102. Extract all the fault codes in the fault tree, and arrange and combine them according to the AND or non-logic between the faults to generate multiple fault samples;

Step 103. Standardize the fault code;

Specifically, when obtaining the data of the fault tree, the fault tree is composed of multiple fault codes with a certain logical relationship. First, it is necessary to extract the fault codes, and then generate multiple The fault tree expands the number of samples to make the diagnosis more accurate.

Further as a preferred embodiment, in the concrete implementation of the present invention, described step 2 is to the following operation steps to the data processing of fault tree:

Step 201. Extract n data samples with the same sample capacity from all the data samples by the method of flush sampling with replacement, as the training samples for training the decision tree;

Step 202. Randomly extract feature variables of m from a training sample;

Step 203. At the internal nodes of the decision tree, select a feature Xi with the best classification effect from the m feature variables according to the principle of the minimum Gini impurity, and divide the node into two branches. The Gini impurity principle is Where P(i) represents the proportion of each category to the total number of categories;

Step 204. Repeat the operation described in step 203 for each internal node of the decision tree until the decision tree can accurately classify the training samples or the Gini impurity of each node of the decision tree reaches the minimum;

Step 205. Select the next training sample, repeat step 202 to step 204, until the decision tree corresponding to the purchased training sample is constructed;

Step 206. The decision trees constructed from the n training samples together form a random forest model, and the training of the random forest model is completed.

Specifically, based on each training sample, the present invention constructs a decision tree for each sample in a completely split manner, and does not perform pruning operations on the constructed decision tree, so that the decision tree can reach the requirements of low deviation and high difference Fully grow to improve the decision tree function, and finally improve the judgment accuracy of the random forest model.

Claims (2)

1. a kind of winged control fault detection and diagnosis method based on random forest, which comprises the following steps:

Step 1. analyzes all true faults that flight control system is included, error code and error code corresponding to every kind of fault type Between connection；

Step 2. is using known fault type as data sample, using error code corresponding to fault type as data sample Characteristic variable completes the training of multiple decision trees using true fault corresponding to each fault type as desired output, by institute The multiple decision trees composition random forest stated:

Step 3. obtains the fault list that current flight control system is produced, and the error code in inventory is input to Random Forest model In, each decision tree exports corresponding prediction result, carries out ballot point to the prediction result of all decision tree output Analysis, Random Forest model final output are that maximum true fault of possibility.

2. a kind of winged control fault detection and diagnosis method based on random forest according to claim 1, which is characterized in that The step S2's method particularly includes:

Step 201. has the consistent data of n sample size of extraction put back to from all data samples in the method for rushing sampling Sample, the training sample as training decision tree；

Step 202. extracts m characteristic variable from a training sample in a random way；

Step 203. chooses one from m characteristic variable at the internal node of decision tree, according to Geordie impurity level minimum principle The node is divided into Liang Ge branch by the best feature Xi of classifying quality, and the Geordie impurity level principle isWherein P (i) indicates the ratio of the total class number of every one kind Zhan；

Step 204. repeats to operate described in step 203 to each internal node of decision tree, until decision tree being capable of Accurate classification The Geordie impurity level of training sample or each node of decision tree reaches minimum；

Step 205. chooses next training sample, repeats step 202 to step 204, until purchasing the training sample pair extracted The decision tree building answered finishes；

The constructed decision tree come out of n training sample described in step 206. collectively constitutes random gloomy neck model, described random gloomy Woods model training finishes.