Fault diagnosis system and method with self-diagnosis function

Abstract

A fault diagnosis system and method having self-diagnosis capabilities is provided. [Solution] The fault diagnosis system includes a plurality of sensors that output a plurality of real-time sensor data in which first real-time sensor data and second real-time sensor data are defined, a storage that stores second historical sensor data corresponding to the second real-time sensor data, and a processor that is signal-connected to the plurality of sensors and the storage, wherein the processor executes a first AI model that outputs first predicted data based on the second real-time sensor data and a second AI model that outputs second predicted data based on the second historical sensor data and the first predicted data, and determines that the difference between the first real-time sensor data and the first predicted data is within an acceptable range, and if it determines that the difference between the second real-time sensor data and the second predicted data is within an acceptable range, outputs a presentation message indicating that the self-diagnosis has passed.

Description

[Technical Field]

[0001] The present invention relates to a fault diagnosis system and method, and more particularly to a fault diagnosis system and method having a self-diagnosis function. [Background technology]

[0002] With the development of artificial intelligence (AI), conventional technologies have enabled processors to run AI models that receive sensor data from multiple sensors and then perform fault diagnosis based on that sensor data. However, the ability of processors to perform fault diagnosis on AI models in conventional technologies is primarily based on the assumption that the multiple sensors and AI models are all functioning normally. However, as the system is used for a long time, abnormalities (e.g., failures) may occur in the sensors, or the AI ​​model itself may exhibit calculation anomalies. These causes may lead to incorrect fault diagnosis of the AI ​​model. Therefore, even if the AI ​​model's diagnosis results are normal, multiple sensors or the AI ​​model's calculations may actually be abnormal. If subsequent applications are based on these diagnosis results, this could pose a risk. Summary of the Invention [Problem to be solved by the invention]

[0003] In view of the above, the main object of the present invention is to provide a fault diagnosis system and method with a self-diagnosis function, thereby avoiding the problem of subsequent applications being executed based on the diagnosis results even when an abnormality occurs in a sensor or AI model. [Means for solving the problem]

[0004] The fault diagnosis system having a self-diagnosis function of the present invention comprises: a plurality of sensors outputting a plurality of real-time sensor data, each of which defines at least one first real-time sensor data and at least one second real-time sensor data; a storage for storing at least one second historical sensor data corresponding to the at least one second real-time sensor data; a processor signally connected to the plurality of sensors and the storage; The processor: a first artificial intelligence (AI) model that outputs at least one first prediction based on the at least one second real-time sensor data; and a second artificial intelligence (AI) model that outputs at least one second predictive data based on the at least one second historical sensor data and the at least one first predictive data; When the processor determines that a difference between the at least one first real-time sensor data and the at least one first predicted data is within an acceptable range, and when the processor determines that a difference between the at least one second real-time sensor data and the at least one second predicted data is within an acceptable range, the processor outputs an indication message indicating that the self-diagnosis has passed. It is characterized by:

[0005] A method for self-diagnosis of a fault diagnosis system executed by a processor, comprising the steps of: Step A: receiving a plurality of real-time sensor data, wherein at least one first real-time sensor data and at least one second real-time sensor data are defined; Step B: outputting at least one first prediction data by a first AI model based on the at least one second real-time sensor data; Step C: determining whether a difference between the at least one first real-time sensor data and the at least one first predicted data is within an acceptable range; Step D: if a difference between the at least one first real-time sensor data and the at least one first predicted data is within an acceptable range, outputting at least one second predicted data by a second AI model based on the at least one second historical sensor data and the at least one first predicted data; Step E: determining whether a difference between the at least one second real-time sensor data and the at least one second predicted data is within an acceptable range; Step F: if a difference between the at least one second real-time sensor data and the at least one second predicted data is within an acceptable range, the processor outputs an indication message indicating that the self-test has passed. A self-diagnosis method for a fault diagnosis system, comprising:

[0006] Therefore, in the present invention, if the processor determines that the difference between at least one first real-time sensor data and at least one first predicted data is within an acceptable range and that the difference between at least one second real-time sensor data and at least one second predicted data is within an acceptable range, the processor can output a model correction message indicating that the self-diagnosis has passed. On the other hand, if the processor determines that the difference between at least one first real-time sensor data and at least one first predicted data is not within an acceptable range, or if the processor determines that the difference between at least one second real-time sensor data and at least one second predicted data is not within an acceptable range, this indicates that an abnormality has occurred in the multiple sensors, the first AI model, or the second AI model. The processor can respond by issuing a warning message (i.e., a model correction message, a second sensor abnormality message, or a first sensor abnormality message) to achieve the self-diagnosis function of the multiple sensors, the first AI model, and the second AI model. The processor can also take corrective action for the abnormality based on the warning message. [Brief explanation of the drawings]

[0007] [Figure 1] 1 is a block diagram of a fault diagnosis system having a self-diagnosis function according to the present invention; [Figure 2] 1 is a process diagram (1) of the fault diagnosis method with self-diagnosis function of the present invention; [Figure 3] FIG. 2 is a block diagram of a first AI model corresponding to step B in the present invention. [Figure 4] FIG. 10 is a block schematic diagram of a second AI model corresponding to step D in the present invention. [Figure 5] 2 is a process diagram (II) of the fault diagnosis method with self-diagnosis function of the present invention; [Figure 6] FIG. 2 is a block schematic diagram of a first AI model corresponding to step G in the present invention. [Figure 7] 1 is a process diagram (III) of the fault diagnosis method with self-diagnosis function of the present invention; [Figure 8] FIG. 10 is a block schematic diagram of a second AI model corresponding to step N in the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0008] As shown in FIG. 1, the fault diagnosis system with self-diagnosis capabilities of the present invention includes multiple sensors, a storage device 20, and a processor 30. While the present invention is described in the exemplary embodiment as being applied to an on-board fuel cell, the scope of application of the present invention is not limited to this. Any fault diagnosis using sensors and an artificial intelligence (AI) computational model is within the scope of application of the present invention. For example, the present invention can be applied to advanced driver assistance systems (ADAS), predictive maintenance systems for boilers that reduce shutdown frequency, wind power generation systems that simulate changes in wind conditions to improve the power generation efficiency of wind turbines, and methods for estimating material properties. In other words, the present invention can be applied to fields such as the chemical industry, aerospace, heat dissipation design for electronic components, biomedicine, and computer graphics (CG). When the present invention is applied to the aforementioned advanced driver assistance system (ADAS), the multiple sensors can be lane detection sensors, including a camera that captures lane images. In this case, real-time sensor data becomes an image of the current lane, and the AI ​​computational model becomes an AI model for lane recognition. The multiple sensors can also be positioning sensors, including a positioning device. In this case, the real-time sensor data will be the actual value of the Global Navigation Satellite System (GNSS) position signal, and the AI ​​calculation model will be the AI ​​model for recognizing the location information.

[0009] The multiple sensors are installed in or electrically connected to the vehicle's fuel cell power plant 10, sense the operating status of the fuel cell power plant 10, and output real-time sensor data related to the fuel cell power plant 10. For example, the multiple sensors include a gas temperature sensor 101, a hydrogen gas pressure sensor 102, an oxygen gas pressure sensor 103, a cell stack temperature sensor 104, a cell stack output current sensor 105, and a cell stack output voltage sensor 106. Correspondingly, the multiple real-time sensor data output by the multiple sensors are the real-time gas temperature Ta, real-time hydrogen gas pressure Ph, real-time oxygen gas pressure Po, real-time cell stack temperature Tb, real-time cell stack output current Io, and real-time cell stack output voltage Vo of the fuel cell power plant 10, respectively. The installation of the multiple sensors in the fuel cell power plant 10 and the measurement principles are common knowledge in the art, so detailed explanations will be omitted.

[0010] For convenience of explanation, the plurality of sensors includes at least one first sensor 111, and the other sensors in the plurality of sensors, except for the at least one first sensor 111, include at least one second sensor 112. Generally, the number of the at least one second sensor 112 is greater than the number of the at least one first sensor 111, but is not limited to this. In an embodiment of the present invention, the first sensor 111 is a cell stack output voltage sensor 106, and the at least one second sensor 112 includes a gas temperature sensor 101, a hydrogen gas pressure sensor 102, an oxygen gas pressure sensor 103, a cell stack temperature sensor 104, and a cell stack output current sensor 105. Correspondingly, the real-time sensor data output by each first sensor 111 is defined as first real-time sensor data S1 (real-time cell stack output voltage Vo), and the real-time sensor data output by each second sensor 112 is defined as second real-time sensor data S2 (real-time gas temperature Ta, real-time hydrogen gas pressure Ph, real-time oxygen gas pressure Po, real-time cell stack temperature Tb, real-time cell stack output current Io). Note that, below, the time point of each real-time sensor data will be described, but this refers to the real-time sensor data at the current time point, except for specially defined previous and next time points.

[0011] The storage 20 is used to store data, and may be, for example, a hard disk drive (HDD), a solid state drive (SSD), a memory, or a memory card.

[0012] The processor 30 may be a central processing unit (CPU) chip or may further include a graphics processing unit (GPU) chip, and the central processing unit chip and the graphics processing unit chip may cooperate with each other. The processor 30 is signal-connected to the multiple sensors and the storage 20, for example, via signal transmission lines and / or buses. Therefore, the processor 30 can receive real-time sensor data from the multiple sensors and access data from the storage 20, so that the processor 30 can store each real-time sensor data in the storage 20 and read each real-time sensor data from the storage 20. After each real-time sensor data is stored in the storage 20, when the processor 30 receives the latest real-time sensor data at the next time point, the real-time sensor data stored in the storage 20 at the previous time point is defined as historical sensor data.

[0013] The processor 30 executes a first artificial intelligence model (hereinafter referred to as the first AI model 31) and a second artificial intelligence model (hereinafter referred to as the second AI model 32). The program data of the first AI model 31 and the program data of the second AI model 32 are stored in the storage 20 and can be accessed and executed by the processor 30. In an embodiment of the present invention, the first AI model 31 and the second AI model 32 may be neural network models. Note that the present invention relates to the application of AI models, and the principles of AI models can be understood by referring to the following documents, so they will not be described in detail here.

[0014] 1. Wuhan University of Technology, 2011 PhD dissertation: Study on some key issues regarding fault diagnosis and maintenance of vehicular fuel cell systems. Author: Quan Rui.

[0015] 2. Hubei University of Technology, 2021 Master's Thesis: Research on Monitoring and Fault Diagnosis System for Vehicle Fuel Cell System. Author: Xie Song.

[0016] 3. Taiyuan University of Technology, 2021 Master's thesis: Research on the design and fault diagnosis method of safety monitoring system for hydrogen fuel cells. Author: Liu Ao.

[0017] As described above, the real-time sensor data output by each first sensor 111 is defined as first real-time sensor data S1, and the real-time sensor data output by each second sensor 112 is defined as second real-time sensor data S2. In this embodiment of the present invention, the first AI model 31 outputs at least one first prediction data based on at least one second real-time sensor data S2. Here, when training is performed, the first AI model 31 includes, as training samples, multiple sets of data related to gas temperature, hydrogen gas pressure, oxygen gas pressure, cell stack temperature, cell stack output current, and cell stack output voltage that are actually measured and collected from the fuel cell power generation system 10 (the same set of data corresponds to the same point in time). Among these, the same set of gas temperature, hydrogen gas pressure, oxygen gas pressure, cell stack temperature, and cell stack output current is used as the input training sample of the first AI model 31, and the same set of cell stack output voltage is used as the output training sample of the first AI model 31. When the first AI model 31 is trained with multiple sets of data and then operated online, the processor 30 uses the real-time gas temperature Ta, the real-time hydrogen gas pressure Ph, the real-time oxygen gas pressure Po, the real-time cell stack temperature Tb, and the real-time cell stack output current Io as input data for the first AI model 31 and multiple second real-time sensor data S2. In other words, the present invention can define at least one second sensor 112 from multiple sensors using the input data items of the first AI model 31, and correspondingly, the first AI model 31 outputs a predicted cell stack output voltage as first prediction data. That is, the present invention can define at least one first sensor 111 from multiple sensors based on the type of output data of the first AI model 31.

[0018] When the second AI model 32 is trained, its training samples include multiple sets of data related to gas temperature, hydrogen gas pressure, oxygen gas pressure, cell stack temperature, cell stack output current, and cell stack output voltage that are actually measured and collected from the fuel cell power generation system 10 (the same set of data corresponds to the same point in time). In the following description, for data at two adjacent points in time, k represents one set of data at a given point in time, and k-1 represents another set of data at the previous point in time. Therefore, gas temperature (k-1), hydrogen gas pressure (k-1), oxygen gas pressure (k-1), cell stack temperature (k-1), cell stack output current (k-1), and cell stack output voltage (k-1) are used as input training samples for the second AI model 32, and gas temperature (k), hydrogen gas pressure (k), oxygen gas pressure (k), cell stack temperature (k), and cell stack output current (k) are used as output training samples for the second AI model 32. When the second AI model 32 is trained with multiple sets of data and then operated online, the processor 30 uses, as multiple second historical sensor data, the previous real-time gas temperature, the previous real-time hydrogen gas pressure, the previous real-time oxygen gas pressure, the previous real-time cell stack temperature, and the previous real-time cell stack output current read from the storage 20, and uses these second historical sensor data and the real-time cell stack output voltage Vo received from the cell stack output voltage sensor 106 or the predicted cell stack output voltage output by the first AI model 31 as input data for the second AI model 32. Correspondingly, the second AI model 32 outputs, as multiple second prediction data, the predicted gas temperature, the predicted hydrogen gas pressure, the predicted oxygen gas pressure, the predicted cell stack temperature, and the predicted cell stack output current. Therefore, the second AI model 32 can output at least one second prediction data based on at least one second historical sensor data and at least one real-time sensor data or at least one first prediction data.

[0019] The process of the fault diagnosis method with self-diagnosis function of the present invention will be described with reference to FIGS.

[0020] Step A: The processor 30 receives a plurality of real-time sensor data from a plurality of sensors. The plurality of real-time sensor data includes at least one first real-time sensor data S1 and at least one second real-time sensor data S2. In an embodiment of the present invention, the at least one first real-time sensor data S1 is a real-time cell stack output voltage Vo, and the at least one second real-time sensor data S2 includes a real-time gas temperature Ta, a real-time hydrogen gas pressure Ph, a real-time oxygen gas pressure Po, a real-time cell stack temperature Tb, and a real-time cell stack output current Io.

[0021] Step B: As shown in Figure 3, the processor 30 outputs at least one first predicted data S1p based on at least one second real-time sensor data S2 using the first AI model 31. In an embodiment of the present invention, the first predicted data S1p is a predicted cell stack output voltage Vop.

[0022] Step C: The processor 30 determines whether the difference between at least one first real-time sensor data S1 and at least one first predicted data S1p is within an acceptable range. Here, the processor 30 has a threshold value set for each sensor, which is an adjustable default value. If the difference is equal to or less than the threshold value, the difference is determined to be within the acceptable range. Conversely, if the difference exceeds the threshold value, the difference is determined to be unacceptable. In an embodiment of the present invention, the processor 30 determines whether the voltage difference between the real-time cell stack output voltage Vo and the predicted cell stack output voltage Vop is equal to or less than a first threshold value. If the determination result in step C is "YES," the voltage difference is determined to be within the acceptable range, and the process of the present invention proceeds to step D described below.

[0023] 4, the processor 30 outputs at least one second prediction data S2p based on at least one second historical sensor data S2h and at least one first prediction data S1p using the second AI model 32. In an embodiment of the present invention, if the processor 30 determines that the voltage difference between the real-time cell stack output voltage Vo and the predicted cell stack output voltage Vop is equal to or less than the first threshold, the processor 30 outputs, using the second AI model 32, predicted gas temperature Ta′, predicted hydrogen gas pressure Ph′, predicted oxygen gas pressure Po′, predicted cell stack temperature Tb′, and predicted cell stack output current Io′ (i.e., the second prediction data S2p) based on the real-time gas temperature Ta′, real-time hydrogen gas pressure Ph′, real-time oxygen gas pressure Po′, real-time cell stack temperature Tb′, and real-time cell stack output current Io′ (i.e., the second historical sensor data S2h) read from the storage 20.

[0024] Step E: The processor 30 determines whether the difference between at least one second real-time sensor data S2 and at least one second predicted data S2p is within an acceptable range. Here, the processor 30 has a threshold value set for each sensor, and these threshold values ​​are as described in step C. In this embodiment of the present invention, the processor 30 determines whether the temperature difference between the real-time gas temperature Ta and the predicted gas temperature Ta" is equal to or less than a second threshold value, whether the air pressure difference between the real-time hydrogen gas pressure Ph and the predicted hydrogen gas pressure Ph" is equal to or less than a third threshold value, whether the air pressure difference between the real-time oxygen gas pressure Po and the predicted oxygen gas pressure Po" is equal to or less than a fourth threshold value, whether the temperature difference between the real-time cell stack temperature Tb and the predicted cell stack temperature Tb" is equal to or less than a fifth threshold value, and whether the current difference between the real-time cell stack output current Io and the predicted cell stack output current Io" is equal to or less than a sixth threshold value. If the processor 30 determines that all of these differences between the second real-time sensor data S2 and the second predicted data S2p are within their respective acceptable ranges, the process of the present invention proceeds to step F described below.

[0025] Step F: If the processor 30 determines that the difference between at least one second real-time sensor data S2 and at least one second predicted data S2p is within the allowable range, the processor 30 outputs a display message Y1 indicating that the self-diagnosis has passed. In this embodiment of the present invention, since it was determined in the above-mentioned step C that the voltage difference between the real-time cell stack output voltage Vo and the predicted cell stack output voltage Vop is within the allowable range, the actual sensor data of the cell stack output voltage sensor 106 and the predicted data of the first AI model 31 match, and therefore the operating status of the cell stack output voltage sensor 106 and the first AI model 31 is normal. Similarly, since it was determined in the above-mentioned step E that the differences between the second real-time sensor data S2 and the second predicted data S2p were within their respective tolerances, the actual sensor data of the gas temperature sensor 101, hydrogen gas pressure sensor 102, oxygen gas pressure sensor 103, cell stack temperature sensor 104, and cell stack output current sensor 105 match the predicted data of the second AI model 32, and therefore the operating status of the gas temperature sensor 101, hydrogen gas pressure sensor 102, oxygen gas pressure sensor 103, cell stack temperature sensor 104, cell stack output current sensor 105, and second AI model 32 are also normal. In summary, the self-diagnosis pass indicated by the presentation message Y1 means that the operating status of all multiple sensors, the first AI model 31, and the second AI model 32 are all normal.

[0026] To summarize the above, if the processor 30 determines that the difference between at least one first real-time sensor data S1 and at least one first predicted data S1p is within the acceptable range (if the judgment result in step C is "YES"), and if the processor 30 determines that the difference between at least one second real-time sensor data S2 and at least one second predicted data S2p is within the acceptable range (if the judgment result in step E is "YES"), the processor 30 outputs a presentation message Y1 indicating that the self-diagnosis has passed.

[0027] In step E, as shown in Figures 2 and 5, if the processor 30 determines that the difference between the at least one second real-time sensor data and the at least one second predicted data is not within an acceptable range, the process of the present invention proceeds to step G described below.

[0028] Step G: The processor 30 continuously receives multiple real-time sensor data from multiple sensors, respectively. Each real-time sensor data received at a next time point can be defined as the latest real-time sensor data. Therefore, as shown in FIG. 6 , the processor 30 uses the first AI model 31 to output at least one latest first predicted data S1pn based on at least one latest second real-time sensor data S2n, and then proceeds to step H described below. In an embodiment of the present invention, the processor 30 uses the first AI model 31 to output a latest predicted cell stack output voltage Vopn based on the latest real-time gas temperature Tan, latest real-time hydrogen gas pressure Phn, latest real-time oxygen gas pressure Pon, latest real-time cell stack temperature Tbn, and latest real-time cell stack output current Ion received at the next time point.

[0029] Step H: The processor 30 determines whether the difference between at least one latest first predicted data S1pn and at least one latest first real-time sensor data S1n is within an acceptable range. In an embodiment of the present invention, the processor 30 determines whether the voltage difference between the latest predicted cell stack output voltage Vopn and the real-time cell stack output voltage Von received at the next time point is less than or equal to a first threshold value.

[0030] If the determination result in step H is "YES," this indicates that the operating state of the second AI model 32 is abnormal (because the difference described in step E is not within the allowable range). Therefore, the processor 30 outputs a model correction message Y2 indicating that the second AI model 32 is abnormal (step I).

[0031] If the determination result in step H is "NO," this indicates that an abnormality (e.g., a failure) has occurred in at least one second sensor 112. As a result, all of the differences described in steps E and H are no longer within the allowable range. Therefore, processor 30 outputs a second sensor abnormality message Y3 indicating that an abnormality has occurred in at least one second sensor 112 (step J).

[0032] In step C, as shown in Figures 2 and 7, if the processor 30 determines that the difference between at least one first real-time sensor data S1 and at least one first predicted data S1p is not within the acceptable range, i.e., if the judgment result in step C is "NO", the process of the present invention proceeds to steps K and L described below.

[0033] Step K: The processor 30 outputs at least one second predicted data S2p based on the at least one second historical sensor data S2h and the at least one first predicted data S1p using the second AI model 32. For Step K, refer to the above-mentioned Step D, and therefore, a description thereof will not be repeated here.

[0034] Step L: The processor 30 determines whether the difference between the at least one second real-time sensor data S2 and the at least one second predicted data S2p is within an allowable range. For Step L, refer to the above-mentioned Step E, and therefore, the description will not be repeated here.

[0035] If the determination result in step L is "YES," this indicates that an abnormality (e.g., a failure) has occurred in at least one of the first sensors 111. This means that the difference described in step C is no longer within the acceptable range. Therefore, the processor 30 outputs a first sensor abnormality message Y4 indicating that an abnormality has occurred in at least one of the first sensors 111 (step M). In an embodiment of the present invention, this means that an abnormality has occurred in the cell stack output voltage sensor 106. If the determination result in step L is "NO," the process of the present invention proceeds to step N, which will be described below.

[0036] Step N: The processor 30 outputs at least one latest second predicted data S2pn based on the at least one second historical sensor data S2h and the at least one first real-time sensor data S1 using the second AI model 32. As shown in Fig. 8, in an embodiment of the present invention, the at least one second historical sensor data S2h input to the second AI model 32 includes a real-time gas temperature Ta' at a previous point in time, a real-time hydrogen gas pressure Ph' at a previous point in time, a real-time oxygen gas pressure Po' at a previous point in time, a real-time cell stack temperature Tb' at a previous point in time, and a real-time cell stack output current Io' at a previous point in time, and the at least one first real-time sensor data S1 input to the second AI model 32 is a real-time cell stack output voltage Vo. In addition, the at least one latest second predicted data S2pn output by the second AI model 32 includes the latest predicted gas temperature Tan”, the latest predicted hydrogen gas pressure Phn”, the latest predicted oxygen gas pressure Pon”, the latest predicted cell stack temperature Tbn”, and the latest predicted cell stack output current Ion”. Comparing step N (Figure 8) with step D (Figure 4), step N (Figure 8) changes the at least one first predicted data S1p input to the second AI model 32 in step D (Figure 4) to at least one first real-time sensor data S1.

[0037] Step O: The processor 30 determines whether the difference between at least one second real-time sensor data S2 and at least one latest second predicted data S2pn is within an acceptable range. Here, the processor 30 sets each threshold value corresponding to each sensor as described in step C. In an embodiment of the present invention, the processor 30 determines whether the temperature difference between the real-time gas temperature Ta and the latest predicted gas temperature Tan″ is equal to or less than a second threshold value, whether the air pressure difference between the real-time hydrogen gas pressure Ph and the latest predicted hydrogen gas pressure Phn″ is equal to or less than a third threshold value, whether the air pressure difference between the real-time oxygen gas pressure Po and the latest predicted oxygen gas pressure Pon″ is equal to or less than a fourth threshold value, whether the temperature difference between the real-time cell stack temperature Tb and the latest predicted cell stack temperature Tbn″ is equal to or less than a fifth threshold value, and whether the current difference between the real-time cell stack output current Io and the latest predicted cell stack output current Ion″ is equal to or less than a sixth threshold value.

[0038] If the determination result in step O is "YES," the processor 30 outputs a model correction message Y2 indicating that an abnormality has occurred in the first AI model 31 (step P). In an embodiment of the present invention, if the differences between these second real-time sensor data S2 and these latest second prediction data S2pn are all within their respective tolerance ranges, this indicates that the operating status of the second AI model 32 is normal, but that an abnormality has occurred in the first AI model 31. This is because the determination result in step C is "NO," and further, based on the cross-checks in steps L and O, it can be determined that an abnormality has occurred in the first AI model 31.

[0039] If the determination result in step O is "NO," the processor 30 outputs a model correction message Y2 indicating that an abnormality has occurred in both the first AI model 31 and the second AI model 32 (step Q). In an embodiment of the present invention, if the differences between these second real-time sensor data S2 and these latest second prediction data S2pn are not all within their respective tolerance ranges, the determination result in step C is "NO." Furthermore, by cross-checking steps L and O, the processor 30 determines in step O that the difference between at least one piece of second real-time sensor data S2 and at least one piece of latest second prediction data S2pn is not within the tolerance range. Therefore, it can be determined that an abnormality has occurred in both the first AI model 31 and the second AI model 32.

[0040] The prompt message Y1, model correction message Y2, second sensor abnormality message Y3, and first sensor abnormality message Y4 are used for back-end abnormality response measures. For example, as shown in FIG. 1 , the processor 30 is electrically connected to a vehicle display 40, a communication interface 50, and / or an on-board electronic system 60. The display 40 may be a dashboard display or a center console display, the communication interface 50 may be a Bluetooth communication interface, a Wi-Fi communication interface, or a mobile communication interface, and the on-board electronic system 60 may be the vehicle's braking system or a system for controlling vehicle speed and direction. The processor 30 may store the prompt message Y1, model correction message Y2, second sensor abnormality message Y3, and first sensor abnormality message Y4 in the storage 20, display them on the display 40 using a light or screen, or transmit them via the communication interface 50 to a smartphone or a cloud server connected to the communication interface 50. Based on the model correction message Y2, the second sensor abnormality message Y3, and the first sensor abnormality message Y4, the processor 30 can control the onboard electronic system 60 and intervene in driving operations, for example, by slowing down or parking on the side of the road to avoid continuing driving despite the possibility of an abnormality in multiple sensors, the first AI model 31, and / or the second AI model 32. [Explanation of symbols]

[0041] 10. Fuel cell power generation equipment 101 Gas temperature sensor 102 Hydrogen gas pressure sensor 103 Oxygen gas pressure sensor 104 Cell stack temperature sensor 105 Cell stack output current sensor 106 Cell stack output voltage sensor 111 First Sensor 112 Second Sensor 20. Storage 30 processors 31 First AI Model 32 Second AI Model 40 Display 50 Communication Interface 60 Automotive Electronic Systems Ta real-time gas temperature pH real-time hydrogen gas pressure Po Real-time oxygen gas pressure Tb Real-time cell stack temperature Io Real-time cell stack output current Vo Real-time cell stack output voltage Vop Estimated cell stack output voltage Real-time gas temperature at the previous time Ta' Real-time hydrogen gas pressure at the previous time point, Ph' Real-time oxygen gas pressure before Po' Tb' Real-time cell stack temperature at the previous time Io' Real-time cell stack output current at the previous point in time Ta” predicted gas temperature Ph” Predicted hydrogen gas pressure Po” Estimated oxygen gas pressure Tb” predicted cell stack temperature Io” predicted cell stack output current S1 First real-time sensor data S1p first prediction data S1pn Latest first forecast data S2 Second real-time sensor data S2p Second Prediction Data S2n Latest second real-time sensor data S2pn Latest second forecast data Tan Latest real-time gas temperature Phn Latest real-time hydrogen gas pressure Latest real-time oxygen gas pressure (Pon) Tbn Latest real-time cell stack temperature Ion Latest real-time cell stack output current Vopn Latest predicted cell stack output voltage Tan” Latest predicted gas temperature Phn” Latest predicted hydrogen gas pressure Latest predicted oxygen gas pressure (Pon) Tbn” Latest predicted cell stack temperature Ion” Latest predicted cell stack output current S2h Second historical sensor data Y1 prompt message Y2 Model Correction Message Y3 Second sensor abnormality message Y4 1st sensor abnormality message

Claims

1. a plurality of sensors outputting a plurality of real-time sensor data, wherein at least one first real-time sensor data and at least one second real-time sensor data are defined; a storage for storing at least one second historical sensor data corresponding to the at least one second real-time sensor data; a processor signally connected to the plurality of sensors and the storage; The processor: a first artificial intelligence (AI) model that outputs at least one first prediction based on the at least one second real-time sensor data; and a second artificial intelligence (AI) model that outputs at least one second predictive data based on the at least one second historical sensor data and the at least one first predictive data; When the processor determines that a difference between the at least one first real-time sensor data and the at least one first predicted data is within an acceptable range, and when the processor determines that a difference between the at least one second real-time sensor data and the at least one second predicted data is within an acceptable range, the processor outputs a presentation message indicating that the self-diagnosis has passed. A fault diagnosis system having a self-diagnosis function.

2. If the processor determines that the difference between the at least one first real-time sensor data and the at least one first predicted data is within an acceptable range, but determines that the difference between the at least one second real-time sensor data and the at least one second predicted data is not within an acceptable range, the processor outputs at least one latest first predicted data based on the at least one latest second real-time sensor data using the first AI model, and determines whether the difference between the at least one latest first predicted data and the at least one latest first real-time sensor data is within an acceptable range; If it is determined that the difference is within the allowable range, the processor outputs a model correction message indicating that the second AI model is abnormal; If it is determined that the second sensor is not within the acceptable range, the processor outputs a second sensor abnormality message.

2. A fault diagnosis system having a self-diagnosis function according to claim 1.

3. When the processor determines that a difference between the at least one first real-time sensor data and the at least one first predicted data is not within an acceptable range, and when the processor determines that a difference between the at least one second real-time sensor data and the at least one second predicted data is within an acceptable range, the processor outputs a first sensor abnormality message.

3. A fault diagnosis system having a self-diagnosis function according to claim 2.

4. When the processor determines that a difference between the at least one first real-time sensor data and the at least one first predicted data is not within an acceptable range and determines that a difference between the at least one second real-time sensor data and the at least one second predicted data is not within an acceptable range, the processor outputs at least one latest second predicted data based on the at least one second historical sensor data and the at least one first real-time sensor data using the second AI model, and determines whether a difference between the at least one latest second predicted data and the at least one second real-time sensor data is within an acceptable range; If it is determined that the result is within the allowable range, the processor outputs the model correction message indicating that the first AI model is abnormal; If it is determined that the values ​​are not within the allowable range, the processor outputs the model correction message indicating that the first AI model and the second AI model are abnormal.

4. A fault diagnosis system having a self-diagnosis function according to claim 3.

5. A method for self-diagnosis of a fault diagnosis system executed by a processor, comprising the steps of: Step A: receiving a plurality of real-time sensor data, in which at least one first real-time sensor data and at least one second real-time sensor data are defined; Step B: outputting at least one first prediction data by a first AI model based on the at least one second real-time sensor data; Step C: determining whether a difference between the at least one first real-time sensor data and the at least one first predicted data is within an acceptable range; Step D: if a difference between the at least one first real-time sensor data and the at least one first predicted data is within an acceptable range, outputting at least one second predicted data by a second AI model based on the at least one second historical sensor data and the at least one first predicted data; Step E: determining whether a difference between the at least one second real-time sensor data and the at least one second predicted data is within an acceptable range; Step F: if a difference between the at least one second real-time sensor data and the at least one second predicted data is within an acceptable range, the processor outputs a presentation message indicating that the self-diagnosis has passed; A self-diagnosis method for a fault diagnosis system.

6. In step E, if the processor determines that the difference between the at least one second real-time sensor data and the at least one second predicted data is not within an allowable range, the processor outputs at least one latest first predicted data based on the at least one latest second real-time sensor data using the first AI model, and determines whether the difference between the at least one latest first predicted data and the at least one latest first real-time sensor data is within an allowable range; If it is determined that the difference is within the allowable range, the processor outputs a model correction message indicating that the second AI model is abnormal; If it is determined that the second sensor is not within the acceptable range, the processor outputs a second sensor abnormality message.

6. The self-diagnosis method for a fault diagnosis system according to claim 5.

7. In step C, if the processor determines that the difference between the at least one first real-time sensor data and the at least one first predicted data is not within an acceptable range, the processor further performs the following: outputting, by the second AI model, the at least one second predicted data based on the at least one second historical sensor data and the at least one first predicted data; determining whether a difference between the at least one second real-time sensor data and the at least one second predicted data is within an acceptable range, and if determined to be within an acceptable range, the processor outputs a first sensor abnormality message; 7. The self-diagnosis method for a fault diagnosis system according to claim 6.

8. If the processor determines that a difference between the at least one second real-time sensor data and the at least one second forecast data is not within an acceptable range, the processor performs the following: outputting at least one latest second forecast data based on the at least one second historical sensor data and the at least one first real-time sensor data by the second AI model; determining whether a difference between the at least one most recent second forecast data and the at least one second real-time sensor data is within an acceptable range; If it is determined that the result is within the allowable range, the processor outputs the model correction message indicating that the first AI model is abnormal; If it is determined that the values ​​are not within the allowable range, the processor outputs the model correction message indicating that the first AI model and the second AI model are abnormal.

8. The self-diagnosis method for a fault diagnosis system according to claim 7.

Images