Intelligent System Failure Analysis Method and System for Star Catalog Based on Semi-Supervised Denoising Autoencoder

Abstract

The present invention discloses a method and system for failure analysis of a star catalog intelligent system based on a semi-supervised denoising autoencoder, which relates to the technical field of fault detection. The present invention includes: constructing a defect fault library of the star catalog intelligent system according to the causes of the faults of the star catalog intelligent system, and performing ont_hot encoding on the fault states in the defect fault library; receiving the operation data of the star catalog intelligent system in different states. The present invention combines the advantages of supervised training and unsupervised training, utilizes the accuracy of labeled data, the richness and low cost of unlabeled data, and uses a large amount of unlabeled data to improve the accuracy and generalization ability of the model. By corrupting the original input data, supervised training can complete a more robust feature expression for non-linear data, and the adopted Focal loss function can avoid the trained model from shifting to the class with more samples due to the imbalance between samples, making the model pay more attention to the class with fewer samples.

Description

Technical Field

[0001] The present invention relates to the technical field of fault detection, and particularly to a method and system for failure analysis of a star catalog intelligent system based on a semi-supervised denoising autoencoder. Background Art

[0002] With the increasing demand for intelligent autonomy in star catalog tasks, the application of artificial intelligence methods in star catalog operation spacecraft has become an inevitable trend. Whether it is the target recognition task during movement or the pose estimation task during operation control, it simulates the star catalog operation process with functions similar to the human brain, such as perception, reasoning, judgment, learning, and abstraction. This process has strong autonomy and can perform probabilistic reasoning under a given set of input conditions, generating a series of unexpected behaviors. Its generalization and evolution capabilities enable it to abstract the internal laws from complex data and have the characteristics of "growing".

[0003] However, precisely because of these non-deterministic characteristics of artificial intelligence, it also brings significant failure risks to applications, specifically manifested in the following two points: (1) The autonomy and learning characteristics of intelligent technology itself make it more uncontrollable than other technologies. Facing a completely unknown scenario, it is impossible to predict the output results of intelligent algorithms in advance; (2) Risks caused by the imperfection of intelligent technology itself, such as the inexplicability in the decision-making process, being easily interfered, and learning unexpected decision-making routes, etc., may all lead to the task deviating from the original established goal. Based on this, the failure mechanism of the star catalog intelligent model is an urgent problem to be solved. Summary of the Invention

[0004] The purpose of the present invention is to provide a method and system for failure analysis of a star catalog intelligent system based on a semi-supervised denoising autoencoder, which can solve the problem of unclear association between the cause of failure and multi-dimensional data for possible software and algorithm failures in the use of the star catalog intelligent autonomous system.

[0005] According to the first aspect of the present invention, to achieve the above object, the present invention provides the following technical solution: A method for failure analysis of a star catalog intelligent system based on a semi-supervised denoising autoencoder, including the following steps:

[0006] Construct a defect fault library of the star catalog intelligent system according to the cause of the star catalog intelligent system failure, and perform ont_hot encoding on the fault states in the defect fault library;

[0007] Receive the operation data of the star catalog intelligent system in different states, and preprocess the operation data;

[0008] Construct a sample data set by combining the preprocessed data with the corresponding operation state encoding, and divide the sample data set into a training set and a test set;

[0009] Construct a semi-supervised denoising autoencoder, which includes multiple stacked sparse autoencoders, and introduce a Softmax classifier in the last layer of the semi-supervised denoising autoencoder;

[0010] Input the training set data into the semi-supervised denoising autoencoder to complete the optimization and determination of the model parameters, and obtain the trained semi-supervised denoising autoencoder, where the training process includes two parts: unsupervised pre-training and supervised fine-tuning;

[0011] Input the test set data into the trained semi-supervised denoising autoencoder to realize the online evaluation of the system failure reasons, and calculate the precision rate and recall rate indicators of the model for the test set diagnosis.

[0012] Furthermore, the operation data includes ambient temperature, board temperature, board voltage, board current, program running time, number of correctly executed instructions, error of the robotic arm from the target position when executing instructions, confidence level and accuracy rate of target detection. The operation data are all collected during the test of the star catalog intelligent system.

[0013] Furthermore, preprocess the operation data, specifically including the following methods:

[0014] (31) Fill in missing data values: Detect whether the data value is empty. If the proportion of missing data is small, ignore the missing data; if the proportion of missing data is large, supplement the missing data by filling in the average value, median or mode of the missing column data;

[0015] (32) Eliminate abnormal data values: Judge abnormal values, and the detected abnormal data are regarded as abnormal values and deleted;

[0016] (33) Delete redundant data: Calculate the similarity of the data. When the similarity exceeds the threshold, it is considered a duplicate record and deleted;

[0017] (34) Data standardization: Use the Max-Min method for normalization processing, specifically as follows:

[0018]

[0019] In the formula is the new data value, is the original data value, is the minimum value of all original data in the dataset, is the maximum value of all original data in the dataset;

[0020] (35) Data dimensionality reduction: Use the principal component analysis method to transform the linearly correlated n-dimensional features into linearly independent k-dimensional comprehensive features through linear transformation.

[0021] Furthermore, a semi-supervised denoising autoencoder is constructed as follows:

[0022] (41) Construct a traditional autoencoder model, which includes an input layer, a hidden layer, and an output layer. The number of nodes in the input layer is determined according to the number of test data, and the number of nodes in the hidden layer and the output layer is determined according to the dimension of the feature vector;

[0023] (42) Construct a sparse autoencoder by adding a penalty factor term to the loss function of the traditional autoencoder model to impose sparsity constraints on the traditional autoencoder model, thereby forming a sparse autoencoder. The cost function of the sparse autoencoder is:

[0024]

[0025]

[0026]

[0027] In the formula, is the mean square error, is the improved mean square error, is the sparse penalty term coefficient, set to 0.3, is the number of neurons in the hidden layer, is the Kullback-Leibler divergence, is the neuron in the hidden layer The average activation degree for all training data, is the sparsity parameter, set to 0.05 or 0.1, is the true label of the i-th sample, is the output value of the model, is the weight matrix, is the bias vector;

[0028] (43) Stack the sparse autoencoders in a stack structure, and the output of the previous layer network is used as the input of the next layer network, thereby constructing a stacked sparse autoencoder;

[0029] (44) On the basis of the stacked sparse autoencoder, add noise to the original input data, and input the changed data into the stacked sparse autoencoder to obtain a semi-supervised denoising autoencoder.

[0030] Furthermore, there are two ways to add noise to the original input data, as follows:

[0031] (51) Add a Gaussian white noise, and the formula is as follows:

[0032]

[0033] In the formula: is the data after adding noise; is the original data; is the coefficient; is a random number that follows a normal distribution with a mean of 0 and a variance of 1;

[0034] (52) Randomly assign a part of the components in the input vector to 0 with a certain probability.

[0035] Further, input the training set data into the semi-supervised denoising autoencoder to complete the optimization and determination of the model parameters, and obtain the trained semi-supervised denoising autoencoder. The training process includes two parts: unsupervised pre-training and supervised fine-tuning, which are as follows:

[0036] (61) Set the parameters of the semi-supervised denoising autoencoder, including the number of hidden layers and the number of neurons in each layer;

[0037] Set 2 - 5 hidden layers. The number of neurons in the hidden layer is calculated according to the following formula, and the number of neurons in each hidden layer is the same:

[0038]

[0039] In the formula: is the number of neurons in the hidden layer, is the number of samples in the test set, is the number of input neurons, is the number of output neurons; is an arbitrary variable, and its specific value range is from 1 to 5;

[0040] (62) Conduct unsupervised pre-training for each sparse autoencoder. Use the unlabeled data samples and the loss function in step (42). Adopt the layer-by-layer greedy training method and the backpropagation algorithm to train the network parameters of each layer of the semi-supervised denoising autoencoder in turn;

[0041] (63) In the supervised fine-tuning stage, stack multiple supervised sparse autoencoders in a stack structure to form a semi-supervised denoising autoencoder, that is, the hidden layer features of the previous supervised sparse autoencoder are used as the input information of the next one. Remove the decoding layer of the semi-supervised denoising autoencoder and add a Softmax classification layer. Use the Focal loss function and use the backpropagation algorithm to optimize the network parameters of each layer. The Focal loss function is expressed as follows:

[0042]

[0043] In the formula: F represents the Focal loss function, is the balance parameter, is the focusing parameter, For predicting label probabilities.

[0044] According to a second aspect of the present invention, the present invention provides a star catalog intelligent system failure analysis system for a semi-supervised denoising autoencoder, which is used to implement the above-mentioned star catalog intelligent system failure analysis method for a semi-supervised denoising autoencoder, and includes:

[0045] A defect fault library construction module, which is used to construct a defect fault library of the star catalog intelligent system according to the causes of the star catalog intelligent system faults, and perform ont_hot encoding on the fault states in the defect fault library;

[0046] A data preprocessing module, which is used to receive the operation data of the star catalog intelligent system in different states and preprocess the operation data;

[0047] A dataset partitioning module, which is used to form a sample dataset by combining the preprocessed data with the corresponding operation state encoding, and partition the sample dataset into a training set and a test set;

[0048] A construction module, which is used to construct a semi-supervised denoising autoencoder. The fault diagnosis model includes multiple layers of stacked sparse autoencoders, and a Softmax classifier is introduced in the last layer of the fault diagnosis model;

[0049] A training module, which is used to input the training set data into the semi-supervised denoising autoencoder, complete the optimization and determination of the model parameters, and obtain the trained semi-supervised denoising autoencoder, where the training process includes two parts: unsupervised pre-training and supervised fine-tuning;

[0050] A test output module, which is used to input the test set data into the trained semi-supervised denoising autoencoder, realize the online evaluation of the system failure cause, and calculate the precision rate and recall rate indicators of the model for the test set diagnosis.

[0051] According to a third aspect of the present invention, the present invention provides a terminal device, including a memory, a processor, and a computer program stored in the memory and capable of running on the processor. The computer program stored in the memory can run on the processor. When the processor loads and executes the computer program, the above-mentioned star catalog intelligent system failure analysis method for a semi-supervised denoising autoencoder is adopted.

[0052] According to a fourth aspect of the present invention, the present invention provides a storage medium containing computer-executable instructions, and the computer-executable instructions are used to execute the above-mentioned star catalog intelligent system failure analysis method for a semi-supervised denoising autoencoder when executed by a computer processor.

[0053] According to a fifth aspect of the present invention, there is provided a computer program product, which includes a computer program. When the computer program is executed by a processor, it is used to load and execute the above-mentioned method for analyzing the failure of the star catalog intelligent system based on the semi-supervised denoising autoencoder.

[0054] The present invention has at least the following beneficial effects:

[0055] 1. The semi-supervised training adopted by the present invention combines the advantages of supervised training and unsupervised training. By using the accuracy of labeled data, the richness and low cost of unlabeled data, a large amount of unlabeled data is used to improve the accuracy and generalization ability of the model. By corrupting the original input data, supervised training can complete a more robust feature expression for non-linear data and improve the robustness of the model.

[0056] 2. The Focal loss function adopted by the present invention can avoid the deviation of the trained model towards the category with more samples due to the imbalance between samples, enabling the model to pay more attention to the category with fewer samples, thereby improving the accuracy of the model for fault diagnosis of the star catalog intelligent system.

[0057] 3. Introducing Gaussian white noise into the input vector in the present invention helps to prevent model overfitting and can enhance the generalization performance of the model.

[0058] Of course, it is not necessary for any product implementing the present invention to achieve all the above-mentioned advantages simultaneously. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 is a flowchart of the analysis method according to an embodiment of the present invention;

[0060] Figure 2 is a schematic framework diagram of the analysis method according to an embodiment of the present invention;

[0061] Figure 3 is a schematic diagram of the star catalog intelligent system defect and fault library constructed in an embodiment of the present invention;

[0062] Figure 4 is a schematic structural diagram of the autoencoder in an embodiment of the present invention;

[0063] Figure 5 is a schematic structural diagram of the stacked sparse autoencoder in an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0064] Next, the technical solutions in the embodiments of the present disclosure will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present disclosure. Obviously, the described embodiments are only a part of the embodiments of the present disclosure, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present disclosure without creative efforts shall fall within the protection scope of the present disclosure.

[0065] The present invention proposes a method for analyzing the failure of a star catalog intelligent system based on a semi-supervised denoising autoencoder. Aiming at the failure phenomena in the star catalog intelligent system caused by software, algorithms, etc., a data-driven method is used to analyze the reasons for the system failure and determine the key defect factors of the star catalog intelligent system. By collecting the operation data of the star catalog intelligent system in different states during testing, preprocessing the data, and constructing the training set and test set of the model; inputting the training set into the semi-supervised denoising autoencoder (SSDAE) to train the software failure mechanism analysis model of the star catalog intelligent system (the trained semi-supervised denoising autoencoder); using the trained model to analyze the mechanism of software failure, and finally output the analysis result of software failure.

[0066] Please refer to Figure 1 and Figure 2 , the present invention provides a technical solution: a method for analyzing the failure of a star catalog intelligent system based on a semi-supervised denoising autoencoder, including the following steps:

[0067] S1. According to the reasons for the failure of the star catalog intelligent system, construct a defect fault library of the star catalog intelligent system, and perform ont_hot encoding on various operating states (including normal state and various fault states) in the defect fault library;

[0068] As Figure 3 shown, the defect fault library includes fault states caused by software, algorithms, etc. Among them, software defect faults include software design errors, software programming errors, external input errors, exception handling errors, etc.; algorithm defect faults include too small sample data set size, low sample annotation quality, no validation set and test set data, incorrect data preprocessing method, etc.;

[0069] Various fault states such as software defect faults and algorithm defect faults are encoded using ont_hot as shown in the following table. For example, the normal state symbol is N, and the state code is (1,0,0,0,0,0,0,...,0), as shown in Table 1 below:

[0070] Table 1 is the operating state encoding table

[0071] ;

[0072] S2. Receive the operation data of the star catalog intelligent system in different states (including normal state and various fault states), and preprocess the operation data;

[0073] It should be noted that the operation data (raw data) are all collected during the testing process of the star catalog intelligent system. The operation data are data under different operation states, such as parameters like environmental temperature, board temperature, board voltage, board current, program running time, number of correctly executed instructions, and the error of the robotic arm from the target position when executing instructions. These process variable parameters are used as the input of the model after data preprocessing. The preprocessing mainly includes the following methods:

[0074] S21. Filling missing data values: Detect whether the data value is empty. If the proportion of missing data is small, the missing data can be ignored; if it is very important data, the missing data can be supplemented by filling the average value, median, or mode of the missing column data, etc.;

[0075] S22. Removing abnormal data values: Methods such as Z - score, interquartile range, or clustering can be used to judge abnormal values. The detected abnormal data are regarded as abnormal values and deleted;

[0076] S23. Deleting redundant data: Methods such as Euclidean distance and cosine similarity can be used to calculate the similarity of data. When the similarity exceeds a certain threshold, it is considered a duplicate record, and the duplicate record will be deleted;

[0077] S24. Data standardization: In order to eliminate the influence of the dimension between features and make them comparable, generally Max - Min normalization is adopted, which is achieved through the following formula:

[0078]

[0079] In the formula is the new data value, is the original data value, is the minimum value of all original data in the dataset, is the maximum value of all original data in the dataset;

[0080] S25. Data dimensionality reduction: The principal component analysis method can be used. Through linear transformation, the linearly correlated n - dimensional features are transformed into linearly independent k - dimensional comprehensive features, and the information represented by the linearly correlated n - dimensional features is represented by the linearly independent k - dimensional comprehensive features. The linear transformation is the process of projecting the n - dimensional linearly correlated features onto the k - dimensional linearly independent comprehensive features through the linear transformation basis;

[0081] S3. The preprocessed data and the corresponding operation state codes are combined to form a sample dataset, and the sample dataset is divided into a training set and a test set;

[0082] For each type of operating state data, the dataset is divided into a training sample set and a test sample set in a ratio of 7:3. The training sample set is used to train the deep learning model, and the test set is used to test and evaluate the performance of the neural network;

[0083] At this time, the input of the fault diagnosis model is the preprocessed standardized data, and the output is the system operating state code;

[0084] S4. Construct a semi-supervised denoising autoencoder (fault diagnosis model). The semi-supervised denoising autoencoder includes multiple stacked sparse autoencoders, and a Softmax classifier is introduced in the last layer of the semi-supervised denoising autoencoder, specifically as follows:

[0085] S41. Construct a traditional autoencoder model (Autoencoder, AE), which includes an input layer, a hidden layer, and an output layer. The number of nodes in the input layer is determined according to the number of test data, and the number of nodes in the hidden layer and the output layer is determined according to the dimension of the feature vector, as Figure 4 shown;

[0086] Specifically, it includes two stages. The first stage is the encoding stage, which converts the original input data type of the model into the internal data type of the encoder; the second stage is the decoding stage, which converts the internal data type of the encoder into the model output data type. The number of nodes in the input layer is determined according to the number of the test data, and the number of nodes in the hidden layer and the output layer is determined according to the dimension of the feature vector;

[0087] Assume that the input training sample data set of the autoencoder is , where N represents the total number of samples, the number of neurons in the input layer and the output layer is both n, the number of neurons in the hidden layer is m, the input vector expression is , the output vector expression is , the hidden layer vector expression is , where m < n, then the corresponding encoding and decoding processes can be represented by the following formulas respectively:

[0088]

[0089]

[0090] In the formula: represents the encoding weight matrix between the input layer and the hidden layer, represents the encoding bias vector between the input layer and the hidden layer; represents the decoding weight matrix between the hidden layer and the output layer, represents the decoding bias vector between the hidden layer and the output layer; and respectively represent the non - linear activation functions in the encoding and decoding processes; the Relu function is used as the activation function in the construction process. Its advantage is that there is no problem of gradient disappearance, which can ensure that the convergence speed of the auto - encoder model remains in a stable state all the time. Its formula is as follows:

[0091] ;

[0092] S42. Construct a sparse auto - encoder (SAE). Add a penalty factor term to the loss function of the traditional auto - encoder model to impose sparsity constraints on the traditional auto - encoder model, thus forming a sparse auto - encoder (SAE). The cost function of the sparse auto - encoder is:

[0093]

[0094]

[0095]

[0096] In the formula, is the mean square error; is the improved mean square error; is the sparse penalty term coefficient, which can generally be set to 0.3; is the number of neurons in the hidden layer; is the Kullback - Leibler divergence; is the average activation degree of the neurons in the hidden layer for all training data; is the sparsity parameter, which can generally be set to 0.05 or 0.1; is the true label of the i - th sample; is the output value of the model; is the weight matrix; is the bias vector;

[0097] S43. Stack the sparse auto - encoders according to a stack structure, and use the output of the previous - layer network as the input of the next - layer network, thus constructing a stack sparse auto - encoder (SSAE);

[0098] S44. As Figure 5As shown, based on the stacked sparse autoencoder, noise is added to the original input data, and the changed data is input into the stacked sparse autoencoder to obtain a semi-supervised denoising autoencoder (stack sparse denoising auto-encoder, SSDAE), which can also be called a stacked sparse denoising autoencoder. The stacked sparse denoising autoencoder is the model of the semi-supervised denoising autoencoder in the unsupervised pre-training stage;

[0099] There are two ways to add noise to the original input data, which are specifically as follows:

[0100] (S44.1) Add a Gaussian white noise, and the formula is as follows:

[0101]

[0102] In the formula: is the data after adding noise; is the original data; is the coefficient; is a random number that follows a normal distribution with a mean of 0 and a variance of 1;

[0103] (S44.2) Randomly assign a part of the components in the input vector to 0 with a certain probability. SSDAE forces the encoder to learn to extract important features and learn more robust features in the input data, while increasing the generalization ability of the model;

[0104] S5. Input the training set data into the semi-supervised denoising autoencoder to complete the optimization and determination of the model parameters, and obtain the trained semi-supervised denoising autoencoder. The training process includes two parts: unsupervised pre-training and supervised fine-tuning, which are specifically as follows:

[0105] S51. Set the parameters of the semi-supervised denoising autoencoder, including the number of hidden layers and the number of neurons in each layer;

[0106] Set 2 - 5 hidden layers. The number of neurons in the hidden layer is calculated according to the following formula, and the number of neurons in each hidden layer is the same:

[0107]

[0108] In the formula: is the number of neurons in the hidden layer, is the number of samples in the test set, is the number of input neurons, is the number of output neurons; is an arbitrary variable, and the specific value range is from 1 to 5;

[0109] S52. Perform unsupervised pre-training on each sparse autoencoder. Use unlabeled data samples and the loss function in step S42. Adopt a layer-by-layer greedy training method and the backpropagation algorithm to train the network parameters of each layer of the semi-supervised denoising autoencoder in sequence.

[0110] Specifically, the unsupervised process consists of N sparse autoencoders. Each sparse autoencoder includes three layers: an input layer + a hidden layer + an output layer. The unsupervised process is composed of N input layers + hidden layers + output layers.

[0111] S53. In the supervised fine-tuning stage, stack multiple supervised sparse autoencoders in a stack structure to form a semi-supervised denoising autoencoder. That is, the hidden layer features of the previous supervised sparse autoencoder are used as the input information for the next one. Remove the decoding layer of the semi-supervised denoising autoencoder and add a Softmax classification layer. Use the Focal loss function and utilize the backpropagation algorithm to optimize the network parameters of each layer.

[0112] The supervised fine-tuning process is as follows: the first input layer + the first hidden layer + the second input layer + the second hidden layer + the third input layer + the third hidden layer +... the Nth input layer + the Nth hidden layer + the Nth output layer + the softmax classification layer.

[0113] Adopting the Focal loss function can avoid the trained model from shifting towards the class with more samples due to the imbalance between samples (the proportion of normal data samples is large while the proportion of faulty data samples is small). The Focal loss function is expressed as follows:

[0114]

[0115] In the formula: F represents the Focal loss function, representing is the balancing parameter, is the focusing parameter, is the predicted label probability;

[0116] It should be noted that the entire training process is called semi-supervised, that is, the combination of unsupervised pre-training and supervised fine-tuning, with a total of 2 processes. And the model trained by supervised fine-tuning is further improved based on the model used in unsupervised pre-training. After completing the training of the two stages, the semi-supervised denoising autoencoder can be obtained.

[0117] S6. Input the test set data into the semi-supervised denoising autoencoder model to realize the online evaluation of the failure mechanism model of the star catalog intelligent autonomous system, conduct an evaluation of the training results, and further analyze and evaluate the online evaluation results of the model by introducing evaluation indicators such as precision, recall rate, and F1 value.

[0118] In summary, the present invention collects the operation data of the star catalog intelligent system in different states during testing, preprocesses the data such as filling missing values, removing outliers, and standardizing, and at the same time performs ont_hot encoding on the fault types. Each type of operation state data is randomly classified according to a ratio and divided into a training sample set and a test sample set. Two-stage training is adopted. First, each SAE is pre-trained without supervision using unlabeled data samples, and a layer-by-layer greedy training method and backpropagation algorithm are used to train the network parameters of each layer of the SSDAE model in turn. Secondly, in the supervised fine-tuning stage, multiple supervised SAEs are stacked according to a stack structure to form an SSDAE, and a Softmax classification layer is added. The Focal loss function is used, and the backpropagation algorithm is used to optimize the network parameters of each layer. Finally, the test sample data set is input into the semi-supervised denoising autoencoder model to realize the online evaluation of the failure mechanism model of the star catalog intelligent autonomous system and analyze the real-time operation state of the star catalog intelligent system.

[0119] Embodiment 2:

[0120] This embodiment provides a failure analysis system for a star catalog intelligent system based on a semi-supervised denoising autoencoder, which is used to implement the failure analysis method of the star catalog intelligent system based on the semi-supervised denoising autoencoder described in Embodiment 1, and includes:

[0121] A defect fault library construction module, which is used to construct a defect fault library of the star catalog intelligent system according to the causes of the star catalog intelligent system faults, and perform ont_hot encoding on the fault states in the defect fault library;

[0122] A data preprocessing module, which is used to receive the operation data of the star catalog intelligent system in different states and preprocess the operation data;

[0123] A data set division module, which is used to form a sample data set by combining the preprocessed data with the corresponding operation state encoding, and divide the sample data set into a training set and a test set;

[0124] A construction module, which is used to construct a semi-supervised denoising autoencoder. The fault diagnosis model includes multiple stacked sparse autoencoders, and a Softmax classifier is introduced in the last layer of the fault diagnosis model;

[0125] A training module, which is used to input the training set data into the semi-supervised denoising autoencoder to complete the optimization and determination of the model parameters, and obtain the trained semi-supervised denoising autoencoder, where the training process includes two parts: unsupervised pre-training and supervised fine-tuning;

[0126] A test output module, which is used to input the test set data into the trained semi-supervised denoising autoencoder to realize the online evaluation of the system failure cause, and calculate the precision rate and recall rate indicators of the model for the test set diagnosis.

[0127] Specifically, the above-mentioned defect fault library construction module, data preprocessing module, data set division module, construction module, training module and test output module can be embedded in a computer processing system. The computer, based on the above-provided star catalog intelligent system failure analysis method using a semi-supervised denoising autoencoder, calls the above-mentioned modules to complete the task of fault diagnosis for the star catalog intelligent system; the above-mentioned defect fault library construction module, data preprocessing module, data set division module, construction module, training module and test output module can perform operations according to the specific steps given by the above-mentioned star catalog intelligent system failure analysis method using a semi-supervised denoising autoencoder.

[0128] It should be noted that the division of each module of the above system is only a logical function division. In actual implementation, it can be fully or partially integrated into a physical entity, or physically separated. And these modules can all be implemented in the form of software called by processing elements; they can also all be implemented in the form of hardware; or some modules can be implemented in the form of software called by processing elements, and some modules can be implemented in the form of hardware. For example, the shared remote driving system construction module can be a separately established processing element, or can be integrated in a certain chip of the above device. In addition, it can also be stored in the memory of the above device in the form of program code, and the function of the above signal processing module can be called and executed by a certain processing element of the above device. The implementation of other modules is similar. In addition, all or part of these modules can be integrated together or can be independently implemented. The processing element mentioned here can be an integrated circuit with the ability to process signals. In the implementation process, each step of the above method or each of the above modules can be completed by the integrated logic circuit in the processor element or the instruction in the form of software.

[0129] For example, the above-mentioned modules can be configured to implement one or more integrated circuits of the above method, such as: one or more application specific integrated circuits (ASICs), or, one or more digital signal processors (DSPs), or, one or more field programmable gate arrays (FPGAs), etc. Again, when a certain module above is implemented in the form of a processing element scheduling program code, the processing element can be a general-purpose processor, such as a central processing unit (CPU) or other processors that can call program code. Again, these modules can be integrated together and implemented in the form of a system-on-a-chip (SOC).

[0130] Embodiment Three:

[0131] The present invention provides a computer program product, which includes a computer program. When the computer program is executed by a processor, it is used to load and execute the above-mentioned method for analyzing the failure of the star catalog intelligent system of the semi-supervised denoising autoencoder.

[0132] It should be noted that, in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprise", "include" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or further includes elements inherent to such process, method, article or device.

[0133] For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances. When an element is referred to as being "assembled on", "installed on", "fixed to" or "disposed on" another element, it can be directly on the other element or there may also be an intermediate element. When an element is considered to be "connected" to another element, it can be directly connected to the other element or there may be an intermediate element at the same time. The terms "vertical", "horizontal", "upper", "lower", "left", "right" and similar expressions used herein are for illustrative purposes only and do not represent the only implementation.

[0134] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

[0135] In the description of this specification, the descriptions referring to terms such as "one embodiment", "example", "specific example", etc. mean that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present disclosure. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in a suitable manner in any one or more embodiments or examples.

Claims

1. A method for failure analysis of a star-table intelligent system based on a semi-supervised denoising autoencoder, characterized in that: The following steps are involved: According to the causes of the star-table intelligent system failures, a defect fault library of the star-table intelligent system is constructed. The defect fault library includes software defect faults and algorithm defect faults, and the fault states in the defect fault library are encoded ont_hot. Software defect faults include software design errors, software programming errors, external input errors, and exception handling errors. Algorithm defect faults include sample data set size is too small, sample annotation quality is low, there is no verification set and test set data, and preprocessing data method; Receive the operating data of the star catalog intelligent system in different states and pre-process the operating data; The preprocessed data and the corresponding running state codes constitute a sample data set, and the sample data set is divided into a training set and a test set; Construct a semi-supervised denoising autoencoder, which includes a multi-layer stacked sparse autoencoder, and introduces a Softmax classifier in the last layer of the semi-supervised denoising autoencoder. Noise is added to the original input data of the semi-supervised denoising autoencoder. The semi-supervised denoising autoencoder extracts features and learns more robust features in the original input data to increase the generalization ability of the semi-supervised denoising autoencoder. The number of nodes in the input layer of the semi-supervised denoising autoencoder is determined according to the number of test data, and the number of nodes in the hidden layer and the output layer is determined according to the dimension of the feature vector. Input the training set data into the semi-supervised denoising autoencoder to optimize the model parameters and obtain the trained semi-supervised denoising autoencoder. The training process includes two parts: unsupervised pre-training and supervised fine-tuning. First, each SAE is pre-trained in an unsupervised manner. Unlabeled data samples are used, and the greedy training method and back propagation algorithm are adopted to train the network parameters of each layer of the SSDAE model in turn. Secondly, in the supervised fine-tuning stage, multiple supervised SAEs are stacked up according to the stack structure to form SSDAE, and a Softmax classification layer is added. The Focal loss function is used, and the back propagation algorithm is used to optimize the network parameters of each layer. The test set data is input into the trained semi-supervised denoising autoencoder to realize online evaluation of the cause of system failure and calculate the precision and recall rate indicators of the model for the test set diagnosis.

2. The method for failure analysis of a star table intelligent system of a semi-supervised denoising autoencoder according to claim 1, characterized in that: The operating data include ambient temperature, board temperature, board voltage, board current, program running time, number of correct instruction executions, error in the distance from the target position when the robotic arm executes instructions, confidence and accuracy of target detection. The operating data are all collected during the test of the star table intelligent system.

3. The method for failure analysis of a star table intelligent system of a semi-supervised denoising autoencoder according to claim 1, characterized in that: Preprocess the running data, including the following methods: (31) Filling missing data values: Check whether the data value is empty. If the missing data accounts for a small proportion, ignore the missing data; if the missing data accounts for a large proportion, fill the missing data by filling the mean, median or mode of the missing column data; (32) Eliminate abnormal data values: Determine abnormal values, and the detected abnormal data will be regarded as abnormal values ​​and deleted; (33) Delete redundant data: Calculate the similarity of the data. When the similarity exceeds the threshold, it is considered a duplicate record and deleted; (34) Data normalization: The Max-Min method was used for normalization, as follows: Where x′ is the new data value, x is the original data value, and x min is the minimum value of all original data in the data set, x max is the maximum value of all original data in the data set; (35) Data dimensionality reduction: Principal component analysis is used to transform linearly related n-dimensional features into linearly independent k-dimensional comprehensive features through linear transformation.

4. The method for failure analysis of a star table intelligent system of a semi-supervised denoising autoencoder according to claim 3, characterized in that: Construct a semi-supervised denoising autoencoder as follows: (41) Construct a traditional autoencoder model, which includes an input layer, a hidden layer, and an output layer. The number of nodes in the input layer is determined according to the number of test data, and the number of nodes in the hidden layer and the output layer is determined according to the dimension of the feature vector; (42) Construct a sparse autoencoder. Add a penalty factor term to the loss function of the traditional autoencoder model to restrict the sparsity of the traditional autoencoder model, and then construct a sparse autoencoder. The cost function of the sparse autoencoder is: Where J(W, b) is the mean square error; J sparse (W, b) is the improved mean square error; β is the coefficient of sparse penalty term, which is set to 0.3; m is the number of neurons in the hidden layer; is the Kullback-Leibler divergence; is the average activation of hidden layer neuron j for all training data; ρ is the sparsity parameter, set to 0.05 or 0.1; i is the true label of the i-th sample; is the output value of the model; W is the weight matrix; b is the bias vector; (43) The sparse autoencoders are stacked according to a stack structure, and the output of the previous layer of network is used as the input of the next layer of network, thereby constructing a stacked sparse autoencoder; (44) Based on the stacked sparse autoencoder, noise is added to the original input data, and the changed data is input into the stacked sparse autoencoder to obtain a semi-supervised denoising autoencoder.

5. The method for failure analysis of a star table intelligent system of a semi-supervised denoising autoencoder according to claim 4, characterized in that: There are two ways to add noise to the original input data, as follows: (51) Add a Gaussian white noise, the formula is as follows: ε~N(0,1) Where: is the data after adding noise; x is the original data; μ is the coefficient; ε is a random number that follows a normal distribution with a mean of 0 and a variance of 1; (52) Randomly assign a portion of the components in the input vector to 0 according to probability.

6. The method for failure analysis of a star table intelligent system of a semi-supervised denoising autoencoder according to claim 5, characterized in that: The training set data is input into the semi-supervised denoising autoencoder to optimize the model parameters and obtain the trained semi-supervised denoising autoencoder. The training process includes two parts: unsupervised pre-training and supervised fine-tuning, as follows: (61) Setting the parameters of the semi-supervised denoising autoencoder, including the number of hidden layers and the number of neurons in each layer; The hidden layer is set to 2-5 layers, and the number of neurons in the hidden layer is calculated according to the following formula. The number of neurons in each hidden layer is the same: Where: N is the number of neurons in the hidden layer, N test is the number of test set samples, N input is the number of input neurons, N output is the number of output neurons; α is an arbitrary value variable, with a specific value range of 1 to 5; (62) performing unsupervised pre-training on each sparse autoencoder, using unlabeled data samples, and using the loss function in step (42), adopting a layer-by-layer greedy training method and a back propagation algorithm, and training each layer of the network parameters of the semi-supervised denoising autoencoder in turn; (63) In the supervised fine-tuning stage, multiple supervised sparse autoencoders are stacked up in a stacked structure to form a semi-supervised denoising autoencoder. That is, the hidden layer features of the previous supervised sparse autoencoder are used as the input information of the next one. The decoding layer of the semi-supervised denoising autoencoder is removed, and a Softmax classification layer is added. The Focal loss function is used to optimize the network parameters of each layer using the back propagation algorithm. The Focal loss function is expressed as follows: Where: F represents the Focal loss function, α is the balance parameter, γ is the focusing parameter, is the predicted label probability.

7. A semi-supervised denoising autoencoder star table intelligent system failure analysis system, used to implement the semi-supervised denoising autoencoder star table intelligent system failure analysis method according to any one of claims 1 to 6, characterized in that: include: The defect fault library construction module is used to construct the defect fault library of the star table intelligent system according to the cause of the star table intelligent system failure, and to perform ont_hot coding on the fault status in the defect fault library; A data preprocessing module is used to receive the operating data of the star catalog intelligent system in different states and preprocess the operating data; A data set division module is used to form a sample data set by combining the preprocessed data and the corresponding running state code, and divide the sample data set into a training set and a test set; The building module is used to build a semi-supervised denoising autoencoder. The fault diagnosis model includes a multi-layer stacked sparse autoencoder, and the last layer of the fault diagnosis model introduces a Softmax classifier; The training module is used to input the training set data into the semi-supervised denoising autoencoder to optimize the model parameters and obtain the trained semi-supervised denoising autoencoder. The training process includes two parts: unsupervised pre-training and supervised fine-tuning. The test output module is used to input the test set data into the trained semi-supervised denoising autoencoder to achieve online evaluation of the cause of system failure and calculate the model's precision and recall indicators for the test set diagnosis.

8. A computer program product, characterized in that The computer program product comprises a computer program, which, when executed by a processor, is used to load and execute the star-table intelligent system failure analysis method of the semi-supervised denoising autoencoder according to any one of claims 1 to 6.

Images