A multi-source sensor fault detection method and computer device

Abstract

The application relates to the technical field of intelligent monitoring, and discloses a multi-source sensor fault detection method, computer equipment, a computer readable storage medium and a computer program product, to solve the problem that traditional sensor fault detection methods need human intervention and cannot accurately detect, locate and correct faults of a large number of multi-source sensors in a ship power system in real time. The method comprises the following steps: acquiring sensor data collected by sensors of multiple types; identifying faults of the sensors by using a convolutional neural network, to determine the types of the sensor faults; reconstructing sensor data collected by the sensors with faults by using a convolutional autoencoder corresponding to the types of the sensor faults, to output correction data; and replacing the sensor data collected by the sensors with faults with the correction data, to output a reconstruction data matrix. The method can accurately detect, classify and reconstruct fault data of the sensors.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent monitoring, and more particularly to a multi-source sensor fault detection method, a computer device, a computer readable storage medium and a computer program product. BACKGROUND

[0002] With the continuous progress of sensing technology and detection technology, intelligent monitoring systems for monitoring ship power systems have developed rapidly. Accurate sensor data can provide information about the operation of the ship power system to help predict the operating state of the ship and help calculate control parameters for the subsystem.

[0003] However, due to aging, defects or environmental factors of sensors, sensors are prone to faults such as data loss, random noise and drift. Although traditional sensor fault detection methods can detect and isolate faults and reconstruct damaged or lost sensor data, they require a lot of human intervention and cannot accurately detect, locate and correct faults in a large number of multi-source sensors in a ship power system in real time. SUMMARY

[0004] To solve the above problems, the present application provides a multi-source sensor fault detection method, a computer device, a computer readable storage medium and a computer program product, which can detect, classify and reconstruct fault data of sensors in real time, and improve the accuracy of sensor fault detection.

[0005] To achieve the above-mentioned purpose, according to the first aspect of the present application, a multi-source sensor fault detection method is provided, which is applied to a ship power system, comprising:

[0006] Obtaining sensor data collected by sensors of multiple types;

[0007] Identifying faults of the sensors by a convolutional neural network to determine types of the faults of the sensors;

[0008] Reconstructing sensor data collected by the sensors having the faults by a convolutional autoencoder corresponding to the types of the faults of the sensors to output corrected data, the number of the convolutional autoencoders being equal to the number of the types of the faults of the sensors;

[0009] Replacing the sensor data collected by the sensors having the faults with the corrected data to output a reconstructed data matrix.

[0010] Further, obtaining the sensor data collected by the sensors of multiple types comprises obtaining the sensor data collected by the sensors of multiple types through a sliding window with a preset time length, and the sensor data comprises at least one of flow, pressure and temperature of steam and condensate generated in a power conversion process of the ship power system.

[0011] Further, the sensor data collected by the plurality of types of sensors is acquired, including acquiring the sensor data collected by the plurality of types of sensors, and the sensor data is input to the convolutional neural network in a matrix form.

[0012] Further, the sensor data is subjected to fault identification by the convolutional neural network to determine the type of sensor fault, including detecting whether there is abnormal sensor data by the convolutional neural network, and classifying the sensor fault and determining the type of sensor fault in the case where the abnormal sensor data is detected.

[0013] Further, the type of sensor fault includes a data missing fault, a random noise fault and a drift fault, and the sensor fault is classified and the type of sensor fault is determined in the case where the abnormal sensor data is detected, including determining that the sensor fault is the data missing fault in the case where the sensor data is missing data is detected, determining that the sensor fault is the random noise in the case where the sensor data is random data is detected, and determining that the sensor fault is the drift fault in the case where the sensor data is drift data is detected.

[0014] Further, the convolutional neural network includes a first convolutional layer, a first maximum pooling layer, a second convolutional layer, a second maximum pooling layer, a first full connection layer, a second full connection layer and a softmax layer.

[0015] Further, the sensor data collected by the sensor that has failed is replaced by the corrected data to output a reconstructed data matrix, including calculating a difference metric of the sensor data collected by the sensor that has failed and the corrected data, and replacing the sensor data collected by the sensor that has failed by the corrected data to output the reconstructed data matrix in the case where the difference metric is greater than a preset value.

[0016] According to a second aspect of the present application, a computer device is also provided, which includes a memory, a processor and a computer program stored in the memory, and the processor executes the computer program to implement the steps of any of the above methods.

[0017] According to a third aspect of the present application, a computer readable storage medium is also provided, which stores a computer program, and the computer program is executed by a processor to implement the steps of any of the above methods.

[0018] According to a fourth aspect of the present application, a computer program product is also provided, which includes a computer program, and the computer program is executed by a processor to implement the steps of any of the above methods.

[0019] Overall, the above technical solutions conceived by the present application can achieve the following beneficial effects compared with the prior art:

[0020] The application provides a multi-source sensor fault detection method, which utilizes the space-time correlation between sensor data collected by multi-source sensors, performs fault detection on the sensor data through a convolutional neural network, determines a fault type when detecting a fault, then reconstructs fault data through a convolutional autoencoder corresponding to the fault type, replaces the fault data with corrected data obtained through reconstruction, and finally outputs a reconstruction data matrix, so that accurate fault detection, fault classification and fault data reconstruction of one or more sensors in a ship power system are realized in real time, and the purpose of avoiding manual intervention is achieved. BRIEF DESCRIPTION OF DRAWINGS

[0021] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required to be used in the embodiments will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without any creative effort on the basis of these drawings.

[0022] Figure 1 A flowchart of a multi-source sensor fault detection method provided by an embodiment of the present application is shown in the figure.

[0023] Figure 2 A flowchart of a multi-source sensor fault detection method provided by another embodiment of the present application is shown in the figure.

[0024] Figure 3 An internal structure diagram of a computer device provided by the embodiment of the present application is shown in the figure. DETAILED DESCRIPTION

[0025] In order to make the objects, technical solutions and advantages of the present application more clear, the present application will be further described in detail below with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application, and are not used to limit the present application. In addition, the technical features involved in each embodiment of the present application described below can be combined with each other as long as they do not conflict with each other.

[0026] The terms "first", "second", "third", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish different objects, and are not used to describe a specific order. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device including a series of steps or units is not limited to the listed steps or units, but can optionally include steps or units not listed, or can optionally include other steps or units inherent to the process, method, product or device.

[0027] AsFigure 1 As shown, a multi-source sensor fault detection method is provided, which is applied to a ship power system. As a complex power system, the ship power system needs a large number of sensors to detect the structural performance, so as to realize the health monitoring of the ship power system. The method can be executed by a terminal or a server in communication with the terminal through a network. The terminal can be, but is not limited to, various personal computers, notebook computers, smart phones, tablet computers, Internet of Things devices, portable wearable devices, and the like. The server can be a stand-alone server or a server cluster composed of multiple servers. Taking the case that the method is applied to the terminal as an example, the method includes the following steps:

[0028] In step 101, sensor data collected by multiple types of sensors is obtained.

[0029] The sensor data collected by the multiple types of sensors (also referred to as multi-source sensors) has spatio-temporal correlation, that is, has a correlation in time sequence and spatial position. The change of one sensor data over time can be analyzed to obtain the change of another sensor data over time, thereby helping to improve the accuracy of the sensor data.

[0030] The sensor data includes the flow, pressure, temperature, etc. of steam and condensate generated in the power conversion process of the ship power system.

[0031] Exemplarily, the terminal obtains the sensor data collected by the multiple types of sensors through a slidable time window of a preset length (for example, 200 ms), and inputs the sensor data in a matrix form to a convolutional neural network (CNN).

[0032] In step 102, the sensor data is subjected to fault identification by the convolutional neural network, to determine the type of sensor fault.

[0033] The convolutional neural network (CNN) is also referred to as a classifier, which is used to identify the type of sensor fault. The sensor data includes normal data, missing data (referred to as T1), random data (referred to as T2), and drift data (referred to as T3), etc. The type of sensor fault includes data missing fault (referred to as RC1), random noise (referred to as RC2), and drift fault (referred to as RC3).

[0034] Exemplarily, the terminal detects whether there is sensor data anomaly through the convolutional neural network; in the case that the sensor data anomaly is detected, the sensor fault is classified to determine the type of sensor fault; in the case that the sensor data anomaly is not detected (i.e., the sensor data is all normal data), the sensor data is not processed.

[0035] When abnormal sensor data is detected, the sensor fault is classified and the type of sensor fault is determined. For example, if the sensor data is missing (T1), the sensor fault is determined to be a data missing fault (RC1); if the sensor data is random (T2), the sensor fault is determined to be random noise (RC2); and if the sensor data is drifting (T3), the sensor fault is determined to be a drift fault (RC3).

[0036] Step 103: Reconstruct the sensor data collected by the faulty sensor using a convolutional autoencoder corresponding to the type of sensor fault, and output corrected data. The number of convolutional autoencoders is equal to the number of sensor fault types.

[0037] Among them, convolutional autoencoders (CAEs) are used for data reconstruction, and the number of CAEs is consistent with the number of sensor fault types.

[0038] For example, when the type of sensor fault is determined, the terminal reconstructs the erroneous data using a convolutional autoencoder corresponding to that type and outputs corrected data. Since the number of convolutional autoencoders is the same as the number of sensor fault types, with one convolutional autoencoder for each type, scalable data reconstruction can be achieved. Even when the number of faulty sensors is large, parallel fault processing can also be achieved, that is, reconstructing sensor data collected by multiple faulty sensors in parallel.

[0039] Step 104: Replace the sensor data collected by the faulty sensor with the correction data and output the reconstructed data matrix.

[0040] The reconstructed data matrix consists of all the sensor data output by all convolutional autoencoders.

[0041] Because the reconstructing of sensor data from a faulty sensor by a convolutional autoencoder can introduce potential biases into the sensor data from other sensors, a differential metric is used to identify the faulty sensor in order to avoid this. The sensor data from the identified faulty sensor is then replaced only in the final output.

[0042] For example, the terminal calculates the difference between the sensor data collected by the faulty sensor and the correction data. If the difference is greater than a preset value, it determines that the sensor data is the data collected by the faulty sensor. At this time, the sensor data collected by the faulty sensor is replaced with the correction data, and the reconstructed data matrix is ​​output.

[0043] The calculation formula of the difference metric is as follows:

[0044]

[0045] Wherein, input is the sensor data collected by the sensor, and output is the corrected data reconstructed by the convolutional autoencoder.

[0046] In the above multi-source sensor fault detection method, compared with the traditional sensor fault detection method which is difficult to be applied to a large number of sensor fault detection, the spatio-temporal correlation between the sensor data collected by the multi-source sensor is utilized, the convolutional neural network is used for fault detection of the sensor data, the fault type is determined when the fault is detected, then the fault data is reconstructed by the convolutional autoencoder corresponding to the fault type, the fault data is replaced by the corrected data reconstructed, and finally the reconstructed data matrix is output, so that the purpose of real-time automatic accurate fault detection, fault classification and fault data reconstruction of one or more sensors in the ship power system is achieved, and manual intervention is avoided.

[0047] In one embodiment, the convolutional neural network includes two alternating convolutional layers and maximum pooling layers (i.e., a first convolutional layer, a first maximum pooling layer, a second convolutional layer, and a second maximum pooling layer), two fully connected layers (i.e., a first fully connected layer and a second fully connected layer), and a softmax layer, for extracting features across multiple time scales and classifying the type of sensor fault based on the features.

[0048] Specifically, since the sensor fault is mainly captured as a time feature (for example, the data missing fault is zero in the reading direction of time), the kernel size is set to 1 to avoid feature averaging of data along the spatial dimension. After the first convolutional layer, the purpose of introducing the first maximum pooling layer is to retain the most important features and reduce the number of parameters in the network. Then a second convolutional layer is added to extract higher time scale features, which is crucial in some cases, for example, a drift or a sharp fault may not be easily perceived at a smaller time scale, but can be identified at a longer time scale. Using cross-entropy loss as the objective function, the kernel size of the first convolutional layer is optimized to (11x1), the kernel size of the second convolutional layer is optimized to (5x1), and the filter size of the two maximum pooling layers is (2x1). Finally, the fully connected layer and the softmax layer are added to classify using the identified features. The training process of the convolutional neural network aims to minimize the cross-entropy loss function, and the cross-entropy loss function is:

[0049]

[0050] Wherein, N is the number of samples, M is the total number of sensor fault categories, y icrepresents the true label of the ith sample belonging to class m (1 if it belongs, otherwise 0), p ic represents the probability that the convolutional neural network predicts that the ith sample belongs to class m.

[0051] In one embodiment, the plurality of sensor data is input to the convolutional neural network in a matrix form, when the plurality of sensor data is stacked into an array, the sensor data is conceptually similar to a pixel, but has a one-dimensional corresponding to the time evolution. It should be noted that although the common machine learning algorithm has a certain accuracy in the image classification task, it does not fully utilize the neighborhood features of the specific point at multiple scales, and in the present application, the convolutional neural network is trained by jointly processing the plurality of sensor data with spatio-temporal correlation, that is, the input data of the convolutional neural network has spatial and temporal correlation, so as to improve the accuracy of sensor data fault detection and classification.

[0052] In one embodiment, the architecture of all convolutional autoencoders is the same for different types of sensor faults. The size of the convolution kernel is selected by minimizing the loss function. Taking the types of sensor faults including data missing fault (RC1), random noise (RC2) and drift fault (RC3) as an example, the three CAE networks are trained independently, and the training process includes minimizing the mean square error, which is the difference between the true value and the CAE network prediction value. Each CAE network processes a specific type of sensor fault.

[0053] In this embodiment, at least one convolutional autoencoder with the same architecture is used to process the corresponding sensor fault, which can improve the accuracy of reconstructing the sensor data collected by the faulty sensor, and can also reduce the computational cost of the convolutional autoencoder in the training and use stages, compared with using only a single convolutional autoencoder to process all sensor faults.

[0054] In one embodiment, as shown in Figure 2 A multi-source sensor fault detection method based on convolutional neural network and convolutional autoencoder, comprising:

[0055] Step one, using a convolutional neural network (classifier) to identify the sensor data collected by the multi-source sensor, and determining the type of sensor fault;

[0056] Step two, using a convolutional autoencoder (reconstructor) to reconstruct the sensor data collected by the faulty sensor.

[0057] Specifically, step one further comprises a preprocessing step, which includes obtaining real-time sensor data through a sliding window of a preset time length (200 ms) as input to the classifier. Step two uses three separate reconstructors (corresponding to three types of faults: data missing fault RC1, random noise RC2, and drift fault RC3). Step two further comprises a post-processing step for fault isolation, which includes identifying the faulty sensor, determining whether it is a sensor fault or a system fault, and outputting the reconstructed data as the final result.

[0058] As shown in the uppermost row, five sensor data S1, S3, S5, S7, and S9 are obtained, among which S7 is missing data. The classifier detects the fault and classifies the fault type, identifies the data missing fault RC1, and calls the corresponding reconstructor (the reconstructor corresponds to the data missing fault RC1) according to the fault type. Subsequently, the reconstructor reconstructs (corrects) the data of S7, predicts the sensor data, and obtains the reconstructed data; and the normal data S1, S3, S5, and S9 are not processed. The post-processing step identifies the only faulty sensor, determines whether it is a sensor fault or a system fault, and outputs the reconstructed data as the final result, obtaining the new sensor data S7 shown in the lowermost row. Figure 2 Figure 2 As shown in the uppermost row, five sensor data S1, S3, S5, S7, and S9 are obtained, among which S7 is missing data. The classifier detects the fault and classifies the fault type, identifies the data missing fault RC1, and calls the corresponding reconstructor (the reconstructor corresponds to the data missing fault RC1) according to the fault type. Subsequently, the reconstructor reconstructs (corrects) the data of S7, predicts the sensor data, and obtains the reconstructed data; and the normal data S1, S3, S5, and S9 are not processed. The post-processing step identifies the only faulty sensor, determines whether it is a sensor fault or a system fault, and outputs the reconstructed data as the final result, obtaining the new sensor data S7 shown in the lowermost row. Figure 2

[0059] In step two, the number of convolutional autoencoders is consistent with the number of types of sensor faults, and the convolutional autoencoders can be expanded as the number of sensors increases. Therefore, the multi-source sensor fault detection method can be applied to a large number of sensor scenarios and can reconstruct data (correct erroneous data) in the presence of different types of sensor faults, so that the erroneous data can be directly used and human intervention can be reduced. Moreover, since the sensor data is collected through a sliding window, the multi-source sensor fault detection method can process sensor faults in real time.

[0060] The number of sensors that can be applied to the multi-source sensor fault detection method is mainly limited by the computational cost of training the CNN and CAE. Since step one classifies sensor faults, step two only needs r (r is the number of types of sensor faults) CAEs to process r types of faults. Assuming that n, m, and r represent the total number of sensors, the maximum number of faulty sensors, and the number of types of sensor faults, respectively, the traditional sensor fault detection method needs ​​Compared with the multi-source sensor fault detection method provided in the embodiment, the CAE needs longer model training time and larger calculation cost. Therefore, the multi-source sensor fault detection method provided in the embodiment can significantly reduce the calculation cost and can be integrated into a data acquisition system or implemented using edge computing, and is particularly suitable for a ship power system with a large number of sensors.

[0061] The application further provides a computer device, an internal structure diagram of which can be as shown in Figure 3 The computer device includes a processor, a memory, an input / output interface, a communication interface, a display unit and an input device. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface, the display unit and the input device are connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium to run. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with external terminals in a wired or wireless manner. The wireless manner can be implemented through WIFI, a mobile cellular network, NFC (near field communication) or other technologies. The computer program is executed by the processor to implement a multi-source sensor fault detection method. The display unit of the computer device is used to form a visually visible picture, which can be a display screen, a projection device or a virtual reality imaging device. The input device of the computer device can be a touch layer overlaid on the display screen, or a key, trackball or touchpad arranged on the shell of the computer device, or an external keyboard, touchpad or mouse, etc.

[0062] Those skilled in the art can understand that Figure 3 The structure shown in the figure is only a block diagram of part of the structure related to the scheme of the application, and does not constitute a limitation on the computer device to which the scheme of the application is applied. A specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0063] As shown in Figure 3 The application further provides a computer device including a memory, a processor and a computer program stored in the memory. The processor executes the computer program to implement the steps in each of the method embodiments.

[0064] The application further provides a computer readable storage medium, which stores a computer program. The computer program is executed by a processor to implement the steps in the above method embodiments. The computer readable storage medium can include, but is not limited to, any type of disk, including a floppy disk, an optical disk, a DVD, a CD-ROM, a micro drive, a magneto-optical disk, a ROM, a RAM, an EPROM, an EEPROM, a DRAM, a VRAM, a flash memory device, a magnetic card or an optical card, a nano system (including a molecular memory IC), or any type of medium or device suitable for storing instructions and / or data.

[0065] The application further provides a computer program product, which includes a computer program. The computer program is executed by a processor to implement the steps in the above method embodiments.

[0066] It should be noted that, for the above method embodiments, in order to simply describe, each of the above method embodiments is described as a series of action combinations, but those skilled in the art should know that the application is not limited to the action sequence described, because according to the application, some steps can be performed in other sequences or simultaneously. Secondly, those skilled in the art should know that the embodiments described in the specification all belong to preferred embodiments, and the actions and modules involved are not necessarily required by the application.

[0067] In the above embodiments, the description of each embodiment is focused on, and the part not described in detail in a certain embodiment can be referred to the related description of other embodiments.

[0068] The above is only exemplary embodiments of the present disclosure, which cannot limit the scope of the present disclosure. Any equivalent changes and modifications made according to the teachings of the present disclosure are still within the scope of the present disclosure. Those skilled in the art will easily think of the embodiments of the present disclosure after considering the specification and practicing the disclosure herein. The present application is intended to cover any variations, uses or adaptive changes of the present disclosure, which follow the general principles of the present disclosure and include common knowledge or conventional technical means in the technical field not recorded in the present disclosure. The specification and examples are only regarded as exemplary, and the scope and spirit of the present disclosure are defined by the claims.

[0069] The technical features of the above embodiments can be combined arbitrarily. In order to make the description simple, not all possible combinations of the technical features in the above embodiments are described, however, as long as the combinations of the technical features do not exist contradictory, they should be considered as the scope recorded in the specification.

[0070] Those skilled in the art can easily understand that the above description is only the preferred embodiment of the present application, and is not intended to limit the present application. Any modification, equivalent replacement and improvement made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A multi-source sensor fault detection method, applied to a marine propulsion system, characterized in that, include: The system acquires sensor data collected by various types of sensors. The sensor data is input into a convolutional neural network in matrix form. The sensor data have spatiotemporal correlations. The sensor data includes at least one of the following: steam, condensate flow rate, pressure, and temperature generated by the ship's power system during power conversion. The sensor data is used to identify faults and determine the type of sensor fault by using a convolutional neural network. The sensor data collected by the faulty sensor is reconstructed by a convolutional autoencoder corresponding to the type of sensor fault, and the corrected data is output. The number of convolutional autoencoders is the number of types of sensor faults. Calculate the difference metric between the sensor data and the correction data collected by the faulty sensor; If the differential metric is greater than a preset value, the sensor data collected by the faulty sensor will be replaced with the correction data, and a reconstructed data matrix will be output.

2. The method as described in claim 1, characterized in that, The acquisition of sensor data collected by various types of sensors includes: Sensor data collected by various types of sensors can be acquired through a sliding window with a preset time length.

3. The method as described in claim 1, characterized in that, The step of identifying sensor faults by using a convolutional neural network to determine the type of sensor fault includes detecting whether there are abnormal sensor data using the convolutional neural network, classifying the sensor faults when abnormal sensor data is detected, and determining the type of sensor fault.

4. The method as described in claim 3, characterized in that, The types of sensor faults include data loss faults, random noise, and drift faults. The process of classifying sensor faults and determining their types when abnormal sensor data is detected includes: If the sensor data is found to be missing, the sensor fault is determined to be a data loss fault. If the sensor data is detected as random data, the sensor fault is determined to be random noise. If the sensor data is detected as drift data, the sensor fault is determined to be a drift fault.

5. The method as described in claim 1, characterized in that, The convolutional neural network includes a first convolutional layer, a first max pooling layer, a second convolutional layer, a second max pooling layer, a first fully connected layer, a second fully connected layer, and a softmax layer.

6. A computer device, characterized in that, The device includes a memory, a processor, and a computer program stored in the memory, characterized in that the processor executes the computer program to implement the method according to any one of claims 1-5.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method according to any one of claims 1-5.

8. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the method according to any one of claims 1-5.

Images