KR102603642B1 - Fault diagnosis method of fuel cell system and digital twin device performing the same - Google Patents

Abstract

A method for diagnosing failures in a fuel cell system and a digital twin device for performing the same are disclosed. The failure diagnosis method of the fuel cell system includes collecting digital twin data for the fuel cell system and storing it in a preset cloud, performing learning using the digital twin data as learning data, and then responding to the physical sensor of the fuel cell system. Step of generating a digital twin model by calculating virtual sensor data of the virtual sensor, Obtaining physical sensor data from the fuel cell system, Calculating the residual of the acquired physical sensor data and the calculated virtual sensor data A step of generating data, performing learning using the generated residual data as learning data, then classifying the residual data to perform fault classification, and performing fault tolerant control for the fuel cell system according to the performance of fault classification. ) includes the step of performing.

Description

Fault diagnosis method of fuel cell system and digital twin device performing the same}

The present invention relates to a method for diagnosing failures in a fuel cell system and a digital twin device that performs the same.

Proton exchange membrane fuel cells (PEMFC) have received increasing attention in recent years due to their environmental friendliness, high power density and low maintenance requirements. Nonetheless, there are issues and flaws in the dynamic loading process that can affect performance even if they are unsafe. Moreover, fault diagnosis, including fault detection and isolation (FDI) of PEMFC, causes great difficulty in controlling the PEMFC system. Traditional FDI systems are data-driven or based on physical models. Generating the failure conditions needed to learn predictive maintenance algorithms on real machines is often very difficult. Furthermore, physically collected data is often limited by collection conditions and the specific elements recorded. It is noteworthy that a model of the system can be developed in the form of a digital twin with associated real-time data through numerous sensors and actuators.

As a next-generation FDI system, digital twin, a smart manufacturing technology, can not only detect mode failures during PEMFC operation, but also isolate faults. By comparing this digital twin with the physical plant, the obsolescence of critical components can be monitored, and anomalies can be detected and isolated at an early stage. The underlying physics can enable isolation of the cause of anomalous behavior, enabling effective self-healing or maintenance of specific hardware.

Republic of Korea Patent Publication No. 10-1418180 (2014.07.03)

The present invention collects digital twin (DT) data for a fuel cell system and stores it in the cloud, and monitors the fuel cell system using the digital twin data stored in the cloud to detect and isolate failures. The purpose is to provide a fault diagnosis method for a fuel cell system that performs fault tolerant control for the fuel cell system and a digital twin device that performs the same.

According to one aspect of the present invention, a method for diagnosing a failure of a fuel cell system performed by a digital twin device is disclosed.

A failure diagnosis method of a fuel cell system according to an embodiment of the present invention includes collecting digital twin data for the fuel cell system and storing it in a preset cloud, performing learning using the digital twin data as learning data, and then performing learning. , generating a digital twin model by calculating virtual sensor data of a virtual sensor corresponding to a physical sensor of the fuel cell system, acquiring physical sensor data from the fuel cell system, the obtained physical sensor data and the calculation Calculating the residual of the virtual sensor data to generate residual data, performing learning using the generated residual data as learning data, classifying the residual data to perform failure classification, and classifying the failure. It includes performing fault tolerant control on the fuel cell system according to the performance of.

The digital twin data includes sensor data output from various sensors installed in each component of the fuel cell system and control input data input to the fuel cell system for fault tolerance control according to the sensor data.

The step of generating the digital twin model involves setting a plurality of virtual sensors using a first multi-layer perceptron neural network (MLPNN) from the digital twin data and estimating the output of the plurality of virtual sensors. The virtual sensor data is generated, and the first multi-layer perceptron neural network is learned using the NBN (Neuron by Neuron) algorithm.

The step of performing the fault classification includes generating a second multi-layer perceptron neural network using the residual data as learning data, training the generated second multi-layer perceptron neural network using the NBN algorithm, and when the learning is completed, the second multi-layer perceptron neural network is trained. 2 It is determined whether the output data of the multi-layer perceptron neural network exceeds a preset threshold, and if it exceeds, the configuration of the fuel cell system where the sensor is located is detected as a failure.

According to another aspect of the present invention, a digital twin device for diagnosing a failure of a fuel cell system is disclosed.

The digital twin device according to an embodiment of the present invention includes a memory that stores a command and a processor that executes the command, wherein the command collects digital twin data for the fuel cell system and stores it in a preset cloud. Step, performing learning using the digital twin data as learning data, and then generating a digital twin model by calculating virtual sensor data of a virtual sensor corresponding to a physical sensor of the fuel cell system, Obtaining sensor data, generating residual data by calculating the residual of the acquired physical sensor data and the calculated virtual sensor data, performing learning using the generated residual data as training data. , performing fault diagnosis by classifying residual data, and performing fault tolerant control on the fuel cell system according to the fault classification. .

The method of diagnosing a failure of a fuel cell system according to an embodiment of the present invention and the digital twin device that performs the same collect digital twin (DT) data for the fuel cell system, store it in the cloud, and store it in the cloud. By monitoring the fuel cell system using stored digital twin data, detecting and isolating failures, and performing fault tolerant control on the fuel cell system, the effects of the failure can be handled and the fuel cell system operates normally. there is.

1 is a diagram showing the concept of a system in which a digital twin device is implemented according to an embodiment of the present invention.
Figure 2 is a flowchart showing a failure diagnosis method of a fuel cell system performed by a digital twin device according to an embodiment of the present invention.
3 is a diagram showing the output and control input of the fuel cell system.
FIG. 4 is a diagram conceptually showing a failure diagnosis method of a fuel cell system according to an embodiment of the present invention of FIG. 2.
Figure 5 is a diagram showing an example of a failure diagnosis method using learning.
Figure 6 is a diagram schematically illustrating the configuration of a digital twin device according to an embodiment of the present invention.

As used herein, singular expressions include plural expressions unless the context clearly dictates otherwise. In this specification, terms such as “consists of” or “comprises” should not be construed as necessarily including all of the various components or steps described in the specification, and some of the components or steps may be included in the specification. It may not be included, or it should be interpreted as including additional components or steps. In addition, terms such as "... unit" and "module" used in the specification refer to a unit that processes at least one function or operation, which may be implemented as hardware or software, or as a combination of hardware and software. .

Hereinafter, various embodiments of the present invention will be described in detail with reference to the attached drawings.

Figure 1 is a diagram showing the concept of a system in which a digital twin device is implemented according to an embodiment of the present invention.

Referring to FIG. 1, a system in which a digital twin device according to an embodiment of the present invention is implemented may be composed of a virtual layer and a physical layer formed in the cloud.

The Fault Tolerant Controller (FTC) of the physical layer is designed to reduce the risk that the proton exchange membrane fuel cells (PEMFC) system may be destroyed due to failure.

Here, the PEMFC system may experience various failures such as hydrogen circuit failure, air compressor failure, and overall battery failure. For example, hydrogen circuit failures include hydrogen blockage and leakage, and air compressor failures include electrical failures due to short circuits and overvoltages, mechanical failures due to stalling of the crankshaft, and hydraulic failures due to reduced compression effectiveness. , Total battery failures include drying out, flooding, etc.

At this time, data input and output from the PEMFC system is recorded and transmitted to the cloud platform. For example, the cloud platform may be Microsoft Azure cloud, AWS (Amazon Web Services) cloud, etc. provided by cloud platform service providers such as Microsoft, Amazon, etc.

In Figure 1, the PEMFC system is only an example as a failure diagnosis target and is not limited thereto. In other words, the digital twin device according to an embodiment of the present invention can be applied to various types of fuel cell systems as well as proton exchange membrane fuel cell systems. Hereinafter, for convenience of understanding and explanation of the invention, the failure diagnosis target will be collectively referred to as a fuel cell system.

FIG. 2 is a flowchart showing a failure diagnosis method of a fuel cell system performed by a digital twin device according to an embodiment of the present invention, FIG. 3 is a diagram showing the output and control input of the fuel cell system, and FIG. 4 is a diagram of FIG. 2. This is a diagram conceptually showing a failure diagnosis method of a fuel cell system according to an embodiment of the present invention, and Figure 5 is a diagram showing an example of a failure diagnosis method using learning. Hereinafter, a failure diagnosis method of a fuel cell system performed by a digital twin device according to an embodiment of the present invention will be described focusing on FIG. 2, with reference to FIGS. 3 to 5.

In step S210, the digital twin device collects digital twin (DT: hereinafter referred to as DT) data about the fuel cell system and stores it in a preset cloud.

For example, as shown in FIG. 3, the digital twin device uses sensor data output from various sensors installed in each component of the fuel cell system and a fault tolerant controller (FTC) to control fault tolerance according to sensor data. Control input data input to the fuel cell system can be collected, and the collected data can be transmitted to the cloud and stored. Collection of such DT data can be done by the Internet of Things (IoT), which stores and updates in the cloud.

In step S220, the digital twin device creates a DT model through learning using DT data stored in the cloud.

In other words, the digital twin device uses DT data stored in the cloud as learning data to set up multiple virtual sensors corresponding to the actual physical sensors of the fuel cell system, and through this, virtual sensor data is calculated by estimating the output of each virtual sensor. You can create a DT model by doing this.

For example, referring to Figure 4, the digital twin device sets a virtual sensor k (k = 1 to N) using a Multi-Layer Perceptron Neural Network (MLPNN) from DT data stored in the cloud, Virtual sensor data can be generated by estimating the output of sensor k. A multilayer perceptron neural network can receive measurement values as input from all sensors except sensor k. A multilayer perceptron neural network can be learned using the NBN (Neuron by Neuron) algorithm. The NBN algorithm can learn quickly and efficiently compared to the widely used error back-propagation training algorithm.

In step S230, the digital twin device acquires physical sensor data from the fuel cell system.

For example, in FIG. 4 , physical sensor data (1 to _N ) where a failure ((f ₁ to f N ) has occurred may be the output of a physical sensor of the fuel cell system.

In step S240, the digital twin device generates residual data by calculating the residual of the acquired physical sensor data and the calculated virtual sensor data.

In other words, the digital twin device can calculate the deviation between physical sensor data and DT model output using the virtual sensor of the generated DT model and the physical sensor of the actual fuel cell system. Additionally, the digital twin device can investigate and point out the location of the fault through residual calculation.

In step S250, the digital twin device classifies the generated residual data and performs failure classification.

In other words, the digital twin device can perform learning using the generated residual data as learning data, then classify the residual data to detect failures and separate the detected failures from other failures.

For example, conventionally, the focus was on fault detection without pointing out fault location and acceptance for the DT model. A digital twin device according to an embodiment of the present invention can detect a failure when the residual exceeds a carefully selected threshold. Referring to FIG. 5, the digital twin device generates a multi-layer perceptron neural network using residual data as learning data, trains the multi-layer perceptron neural network created using the NBN algorithm, and when learning is completed, the output data of the multi-layer perceptron neural network is stored in advance. It is determined whether the set threshold is exceeded, and if it is exceeded, the configuration of the fuel cell system where the sensor is located can be detected as a failure.

After a fault is successfully detected, a fault isolation procedure can be performed to distinguish a particular fault from other faults.

For example, if a failure occurs in the configuration of the fuel cell system where the ith sensor is located, the ith output of the multilayer perceptron neural network will be greater than the threshold. A single residual data is sufficient to detect a fault, but usually a set of residual data (residual vector) is required for fault isolation. If one failure can be distinguished from another failure using one residual set (residual vector), the failure can be said to be separable using that residual set (residual vector). If all faults can be separated using a set of residuals (residual vector), then the set of residuals can be said to have the necessary separation properties. Through the integration of virtual and physical, services such as real-time monitoring, fault diagnosis and maintenance decision-making can be provided. As a result, the durability and reliability of the fuel cell system can be improved.

In step S260, the digital twin device performs fault tolerant control on the fuel cell system by detecting and isolating a failure of the fuel cell system.

For example, as shown in FIG. 4, the digital twin device can control cooling, hydrogen supply, temperature adjustment, oxygen supply, etc. of the fuel cell system so that the residual falls below the threshold.

To improve the reliability of fuel cell systems, a fault-tolerant control strategy is needed to ensure that the fuel cell system fulfills the application requirements and completes its mission. In this work, virtual fault information is immediately converted into control information. In other words, fault tolerance control is intended to deal with the effects of failure in the fuel cell system.

The failure diagnosis method of the fuel cell system according to the embodiment of the present invention can ensure a high level of accuracy and reliability of the fuel cell system in the event of a failure or abnormality. Fault diagnosis based on DT approach using response time of fault detection speed is relatively fast and robust and can identify multiple faults. Additionally, the failure diagnosis method of a fuel cell system according to an embodiment of the present invention can provide a more active and stable maintenance strategy than the conventional one.

The digital twin device according to an embodiment of the present invention builds a DT model that can update system parameters according to operating procedures. The advantage of the digital twin approach lies mainly in the integration of the virtual and the physical. Under the prerequisite of keeping the physical and virtual fuel cell systems consistent, the created digital twin can fuse signals from various sources, such as simulated data, historical data, and real data, for diagnostic purposes. Of course, the more knowledge and information is integrated, the more effective and reliable the diagnosis. In predictive maintenance scenarios, digital twins can play a powerful role in generating data and combining it with available sensor data to build and validate predictive maintenance algorithms. Additionally, a digital twin can integrate the physical understanding of the system and provide a description of the dynamic behavior of the fuel cell system. And in clouds such as AWS Lambda, fault diagnosis based on new DT approaches can be developed. If a failure occurs, an alarm message can be sent to email or the user terminal through AWS Simple Notification Service (SNS) for notification and maintenance decisions. Additionally, by maintaining real-time synchronization between the virtual model and the physical object and integrating interaction with the environment into the fault detection and diagnosis procedure, the fault diagnosis method of a fuel cell system according to an embodiment of the present invention may be better than a data-based method. You can.

Figure 6 is a diagram schematically illustrating the configuration of a digital twin device according to an embodiment of the present invention.

Referring to FIG. 6, the digital twin device according to an embodiment of the present invention includes a processor 10, a memory 20, a communication unit 30, and an interface unit 40.

The processor 10 may be a CPU or a semiconductor device that executes processing instructions stored in the memory 20.

Memory 20 may include various types of volatile or non-volatile storage media. For example, memory 20 may include ROM, RAM, etc.

For example, the memory 20 may store instructions for performing a failure diagnosis method of a fuel cell system according to an embodiment of the present invention.

The communication unit 30 is a means for transmitting and receiving data with other devices through a communication network.

The interface unit 40 may include a network interface and a user interface for accessing the network.

Meanwhile, the components of the above-described embodiment can be easily understood from a process perspective. In other words, each component can be understood as a separate process. Additionally, the processes of the above-described embodiments can be easily understood from the perspective of the components of the device.

Additionally, the technical contents described above may be implemented in the form of program instructions that can be executed through various computer means and recorded on a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, etc., singly or in combination. Program instructions recorded on the medium may be specially designed and configured for the embodiments or may be known and available to those skilled in the art of computer software. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tapes, optical media such as CD-ROMs and DVDs, and magnetic media such as floptical disks. -Includes optical media (magneto-optical media) and hardware devices specifically configured to store and execute program instructions, such as ROM, RAM, flash memory, etc. Examples of program instructions include machine language code, such as that produced by a compiler, as well as high-level language code that can be executed by a computer using an interpreter, etc. A hardware device may be configured to operate as one or more software modules to perform the operations of the embodiments, and vice versa.

The above-described embodiments of the present invention have been disclosed for illustrative purposes, and those skilled in the art will be able to make various modifications, changes, and additions within the spirit and scope of the present invention, and such modifications, changes, and additions will be possible. should be regarded as falling within the scope of the patent claims below.

10: processor
20: memory
30: Department of Communications
40: interface part

Claims (5)

In the method of diagnosing a fuel cell system failure performed by a digital twin device,
Collecting digital twin data for the fuel cell system and storing it in a preset cloud;
After performing learning using the digital twin data as learning data, generating a digital twin model by calculating virtual sensor data of a virtual sensor corresponding to a physical sensor of the fuel cell system;
Obtaining physical sensor data from the fuel cell system;
generating residual data by calculating a residual of the acquired physical sensor data and the calculated virtual sensor data;
performing learning using the generated residual data as learning data, and then classifying the residual data to perform failure classification; and
Including performing fault tolerant control on the fuel cell system according to the performance of the fault classification,
The step of creating the digital twin model is,
A plurality of virtual sensors using a first multi-layer perceptron neural network (MLPNN) are set from the digital twin data, and the virtual sensor data is generated by estimating the output of the plurality of virtual sensors,
A fault diagnosis method for a fuel cell system, characterized in that the first multi-layer perceptron neural network is learned using the NBN (Neuron by Neuron) algorithm.
According to paragraph 1,
The digital twin data is,
Sensor data output from various sensors installed in each component of the fuel cell system; and
A failure diagnosis method for a fuel cell system, comprising control input data input to the fuel cell system for fault tolerance control according to the sensor data.
delete According to paragraph 1,
The step of performing the fault classification is,
A second multi-layer perceptron neural network is generated using the residual data as learning data, the generated second multi-layer perceptron neural network is trained using the NBN algorithm, and when the learning is completed, the output data of the second multi-layer perceptron neural network is pre-processed. A failure diagnosis method for a fuel cell system, characterized in that it determines whether a set threshold is exceeded and, if it exceeds, detects a failure in the configuration of the fuel cell system where the sensor is located.
In the digital twin device for diagnosing failures in the fuel cell system,
Memory for storing instructions; and
Including a processor that executes the instructions,
The command is:
Collecting digital twin data for the fuel cell system and storing it in a preset cloud;
After performing learning using the digital twin data as learning data, generating a digital twin model by calculating virtual sensor data of a virtual sensor corresponding to a physical sensor of the fuel cell system;
Obtaining physical sensor data from the fuel cell system;
generating residual data by calculating a residual of the acquired physical sensor data and the calculated virtual sensor data;
performing learning using the generated residual data as learning data, and then classifying the residual data to perform failure classification; and
Performing a fault diagnosis method for a fuel cell system, including performing fault tolerant control on the fuel cell system according to the performance of the fault classification,
The step of creating the digital twin model is,
A plurality of virtual sensors using a first multi-layer perceptron neural network (MLPNN) are set from the digital twin data, and the virtual sensor data is generated by estimating the output of the plurality of virtual sensors,
A digital twin device characterized in that the first multi-layer perceptron neural network learns using the NBN (Neuron by Neuron) algorithm.

Images