KR102195266B1 - Fault diagnosis method for plant using sound signal - Google Patents

Abstract

The present invention relates to a technology capable of automatically diagnosing a defect state of a plant facility by determining the degree of belonging for each diagnosis area based on fuzzy inference using an acoustic signal generated during plant operation.
The method for diagnosing a defect of a plant facility using an acoustic signal according to the present invention generates first to third membership functions respectively corresponding to an acoustic signal for each diagnosis state of safety, warning, and danger in the plant defect diagnosis apparatus, The first to third membership functions include a first step of extracting the degree to which the input frequency belongs in the diagnosis area of the corresponding membership function as a membership value, and collecting acoustic signals generated when the plant to be diagnosed is driven by the plant defect diagnosis device. The second step, the third step of obtaining a frequency spectrum by FFT conversion of the acoustic signal collected by the plant fault diagnosis device, and determining a frequency having a preset magnitude value or higher in the frequency spectrum as the input frequency, and in the plant fault diagnosis device Including a fourth step of calculating the first to third membership values by applying the input frequency to the first to third membership functions, respectively, and determining a diagnosis state corresponding to the membership function for which the maximum membership value is calculated as the plant diagnosis state. It characterized in that it is configured to.

Description

Fault diagnosis method for plant using sound signal

The present invention relates to a technology capable of automatically diagnosing a defect state of a plant facility by determining the degree of belonging for each diagnosis state area based on a preset fuzzy reasoning using an acoustic signal generated during plant operation.

In general, a plant has a large number of major facilities in a large space, and because such a plant industry consists of high-temperature and high-pressure processes, in the event of a serious accident such as a fire or explosion, the damage scale is very large and a high risk group. It is classified as an industry.

In order to monitor and control the recent operation status efficiently, measuring instruments (thermometers, pressure gauges, etc.) are installed in the plant process system and facilities, and managers instantly check process operation values and determine whether there is any abnormality, but effective management is performed due to temporal and spatial limitations. It is difficult to expect.

In addition, with the development of measurement and control technology, measuring instruments were replaced with measurement sensors and remote monitoring using a computer instead of manpower monitoring, which significantly reduced the time required to determine the situation, but the process of a large plant with tens of thousands of measuring devices installed. In reality, it is difficult to find out which device provided the cause of the failure through a comprehensive analysis of the changes by checking the changes in the system one by one.

In particular, managers with large plants are making efforts to increase their analysis capabilities by installing automated high-tech monitoring devices and control devices, but they rely on specialized personnel rather than computers for detection of subtle abnormal signs and comprehensive judgment of failure situations.

When an abnormal signal occurs in such an industrial plant, experts identify the cause of the abnormal signal by checking the plant facility diagram (P&ID) one by one.However, due to the nature of the plant equipment, tens of thousands of operating signals are observed and the connection between the signals is analyzed to determine the abnormal situation and It is very difficult and time consuming to find a faulty device.

In addition, since plant equipment classified as a high-risk industry can only be repaired (repaired, replaced) and inspected by skilled experts, it is a reality that a tremendous amount of time and cost is consumed for training skilled experts.

Due to this, in the event of an emergency or disaster due to an abnormal condition of the plant equipment, prompt and accurate measures and responses cannot be made, which in turn causes enormous economic and social loss costs.

1. Korean Patent Publication No. 10-2017-0084955 (A method and apparatus for diagnosing a failure of a wind power generator)

Accordingly, the present invention was created in view of the above circumstances, and performs frequency conversion by collecting acoustic signals generated when the plant is driven, and determining the degree of belonging for each diagnostic area based on fuzzy inference using the converted frequency as an input value. Thus, there is a technical object to provide a method for diagnosing a plant defect using an acoustic signal that enables easy diagnosis of a defect condition of a plant.

상기 수학식에서 a와 b는 각 진단 상태별 음향신호 영역에 해당하는 주파수의 최소값과 최대값이고, x는 입력 주파수임. According to one aspect of the present invention for achieving the above object, the first to third membership functions corresponding to sound signals for each diagnostic state of safety, warning, and danger of a plant in a plant fault diagnosis apparatus are generated and stored. The first step and the second step of collecting the acoustic signal generated when the plant to be diagnosed is running in the plant fault diagnostic device, FFT conversion of the acoustic signal collected in the plant fault diagnostic device to obtain a frequency spectrum, and The third step of determining a frequency having a set magnitude value or more as an input frequency, and by applying the input frequency to the first to third membership functions in the plant fault diagnosis apparatus, respectively, calculating the first to third membership values, and the maximum belonging And a fourth step of determining a diagnosis state corresponding to the calculated membership function as a plant diagnosis state, and in the first step, the first to third membership functions have an input frequency in the diagnosis area of the membership function. A method for diagnosing a defect of a plant facility using an acoustic signal, characterized in that it is generated by the following equation, is provided as a function of outputting the belonging degree as a belonging value.

In the above equation, a and b are the minimum and maximum frequencies corresponding to the acoustic signal region for each diagnosis state, and x is the input frequency.

delete

In addition, the first step is a step of collecting sound signals for each diagnostic state in the plant fault diagnosis apparatus, obtaining a frequency spectrum for each diagnostic state by FFT conversion of the collected sound signals for each diagnostic state, and diagnosing in each frequency spectrum. Setting the frequency domain belonging to each condition, extracting the minimum and maximum values of the frequency domain belonging to each diagnosis status as a fuzzy parameter, and applying the fuzzy parameters for each diagnosis status to the above equations, respectively, to the first to third membership functions There is provided a method for diagnosing a defect of a plant facility using an acoustic signal, comprising the step of generating.

In addition, in the third step, when the number of input frequencies extracted by the preset reference magnitude value exceeds the reference number range, the reference magnitude value is set high in a certain unit, and the number of input frequencies is the reference number range. If it is less than, there is provided a method for diagnosing a defect of a plant facility using an acoustic signal, characterized in that the number of input frequencies is set to satisfy the reference number range by setting the reference magnitude value low in a predetermined unit.

In addition, in the fourth step, the plant fault diagnosis apparatus applies a plurality of input frequencies to the first to third membership functions, respectively, calculates frequency membership values for each input frequency for each membership function, and calculates the average of the frequency membership values. There is provided a method for diagnosing a defect of a plant facility using an acoustic signal, characterized in that the membership value for a corresponding membership function is determined.

According to the present invention, it is possible to automatically diagnose whether a plant has a defect using fuzzy inference based on an acoustic signal generated during plant operation.

In addition, by automatically diagnosing and outputting defects in the plant, it is possible to respond to the defect status of the plant in a timely manner without relying on experienced personnel, thereby minimizing the risk of accidents that may occur.

1 is a diagram schematically showing a configuration of a system for diagnosing defects of a plant facility using an acoustic signal to which the present invention is applied.
2 is a flowchart illustrating a method for diagnosing a defect of a plant facility using an acoustic signal according to the first embodiment of the present invention.
3 is a flowchart for explaining the membership function generation process (ST100) shown in FIG. 2;
4 and 5 are diagrams for explaining a process of generating a membership function shown in FIG. 3;
6 is a view for explaining by illustrating a process of determining the current state of the plant shown in FIG. 2;

Since the embodiments described in the present invention and the configurations shown in the drawings are only preferred embodiments of the present invention, and do not represent all the technical spirit of the present invention, the scope of the present invention is limited to the embodiments and drawings described in the text. Should not be construed as limited by That is, since the embodiments can be variously changed and have various forms, the scope of the present invention should be understood to include equivalents capable of realizing the technical idea. In addition, since the object or effect presented in the present invention does not mean that a specific embodiment should include all of them or only those effects, the scope of the present invention should not be understood as being limited thereto.

All terms used herein have the same meaning as commonly understood by one of ordinary skill in the field to which the present invention belongs, unless otherwise defined. Terms defined in a commonly used dictionary should be construed as having the meaning of the related technology in context, and cannot be interpreted as having an ideal or excessively formal meaning that is not clearly defined in the present invention.

In general, the well-known fuzzy theory makes it possible to deal mathematically with fuzzy information, which is an approximate and inaccurate nature of the real world rather than a traditional logical system.

It is more effective in expressing nature). For example, “big”, “low”, and “small” can systematically process a number of unclear knowledge, information, and logic used by humans using fuzzy theory.

The existing set theory consists of a clear-bounded set with a value of 1 when element x belongs to set A and 0 when it does not. On the other hand, in order to represent an ambiguous set such as “a set of pretty flowers”, which cannot be handled by the existing set concept, a fuzzy set is a set representing the degree to which an element belongs to a set as a value between 0 and 1. It is called (Fuzzy set). At this time, a value between 0 and 1, that is, represented by [0, 1], is called a membership function.

When an arbitrary set X is a set of objects composed of consecutive or discrete elements x, X=x is called the argument set. Fuzzy set A on this discussion set X is a set characterized by the membership function μA, which is μA: X→[0, 1]. Here, the μA(x) value of the fuzzy set is called the membership value in x∈X. This membership value represents the degree to which element x belongs to the fuzzy set A and has a value between 0 and 1. Therefore, the fuzzy set A on the argument set X is represented by a pair of element x and its membership function value.

A system for dealing with ambiguous language based on this fuzzy set is called a fuzzy inference system, and the main feature of this fuzzy inference system is the concept of fuzzy partitioning of given information, which is a numerical value ( Crisp number) is based on a fuzzy set, which contains more information than a simple number.

The present invention is to determine a defect state of a plant facility by analyzing an acoustic signal based on such fuzzy inference. At this time, the plant equipment may be various industrial plants including gas plants.

1 is a diagram schematically showing the configuration of a system for diagnosing defects of a plant facility using an acoustic signal to which the present invention is applied.

Referring to FIG. 1, a system for diagnosing defects in plant facilities using an acoustic signal comprises a plant defect diagnosis apparatus 100 for diagnosing a defect condition in a plant 1 based on an acoustic signal obtained from the plant 1 do.

The plant defect diagnosis apparatus 100 determines each membership function corresponding to an acoustic signal generated in a stable state, a warning state, and a dangerous state for the plant 1 and registers it in the data memory 150 in advance.

Thereafter, the plant defect diagnosis apparatus 100 diagnoses the condition of the sound signal generated when the plant 1 is driven based on a preset membership function.

The plant defect diagnosis apparatus 100 includes an acoustic collection unit 110, a signal conversion unit 120, a defect diagnosis unit 130, a result output unit 140, and a data memory 150.

The sound collection unit 110 may be installed at an appropriate location for collecting sound signals during operation of the plant 1, for example, at one side of the plant 1 or at a location spaced apart from the plant 1 by a predetermined distance. The sound collection unit 110 collects an analog sound signal generated when the plant 1 is driven.

The signal conversion unit 120 converts the sound signal obtained by the sound collection unit 110 into a fast Fourier transform (FFT). In this case, the signal conversion unit 120 extracts the sound signal of a certain frame unit at a certain period and performs FFT conversion of the extracted sound signal. For example, the sound signal received for 1 minute is FFT-converted with a period of 1 hour.

The fault diagnosis unit 130 determines an input frequency to be applied to the membership function from among the FFT-transformed frequencies, and calculates a membership value by applying the corresponding input frequency to the membership function, thereby determining a fault state of the plant.

The diagnosis result output unit 140 displays and outputs a defect diagnosis result for a corresponding plant determined by the defect diagnosis unit 130.

The data memory 150 stores information for plant facility defect diagnosis including a membership function for each diagnosis status corresponding to an acoustic signal and a reference value for defect diagnosis.

Next, a method for diagnosing a defect of a plant facility using an acoustic signal according to a first embodiment of the present invention will be described with reference to the flowcharts shown in FIGS. 2 and 3. FIG. 2 shows the overall operation of the method for diagnosing defects in plant equipment using an acoustic signal according to the present invention, and FIG. 3 shows the membership function generation step (ST100) in FIG. 2 in more detail.

Referring to FIG. 2, the plant defect diagnosis apparatus 100 generates a membership function for each diagnosis state in response to sound signals corresponding to a stable state, a warning state, and a dangerous state for the plant (ST100). That is, the plant fault diagnosis apparatus 100 generates a first membership function corresponding to a stable state, a second membership function corresponding to a warning state, and a third membership function corresponding to a dangerous state, respectively, and the data memory 150 Register at

In the above-described state, the plant defect diagnosis apparatus 100 collects an acoustic signal generated when the diagnosis target plant 1 is driven (ST200).

The plant defect diagnosis apparatus 100 obtains a frequency spectrum corresponding to the collected acoustic signal by FFT-converting the collected acoustic signal (ST300).

The plant fault diagnosis apparatus 100 determines an input frequency to be applied to the membership function in the frequency spectrum (ST400). At this time, when the number of input frequencies extracted by the preset reference magnitude value exceeds the reference number range, the plant defect diagnosis apparatus 100 sets the reference magnitude value high in a certain unit and the number of input frequencies is less than the reference number range. In, the reference magnitude value is set to be low in a certain unit, and the number of input frequencies is set to satisfy the reference number range.

The plant fault diagnosis apparatus 100 calls the first to third membership functions stored in the data memory 150 and applies the input frequency determined in step ST400 to the first to third membership functions, Each of the first to third membership values is calculated. (ST500). At this time, the plant fault diagnosis apparatus 100 calculates a frequency membership value for each input frequency for each membership function by applying a plurality of input frequencies to the first to third membership functions, respectively, and calculates the average of the frequency membership values for the corresponding membership. It is determined by the membership value for the function.

Thereafter, the plant defect diagnosis apparatus 100 determines a diagnosis state corresponding to the membership function obtained by calculating the maximum membership value among the first to third membership values calculated in step ST500 as the plant diagnosis state (ST600). That is, when the third membership value calculated by the third membership function corresponding to the dangerous state is the largest, the current state of the plant is determined as the dangerous state.

Thereafter, the plant defect diagnosis apparatus 100 displays and outputs the plant status information in various forms such as numbers, texts, or images corresponding to the current status determined in step ST600 (ST700). For example, if the plant is determined to be in a dangerous state, it can be displayed as "Danger 90", "Dangerous state" or "Danger image red light".

Meanwhile, in the membership function setting step (ST100), as shown in FIG. 3, sound signals for a stable state, a warning state, and a dangerous state generated when the plant is driven are respectively collected (ST110). At this time, the sound signal for each diagnosis state may be provided from an external organization such as a plant manufacturer or a plant-related company, or sound generated from a plant that is actually driven in each state may be collected.

The plant defect diagnosis apparatus 100 obtains a frequency spectrum by performing FFT conversion of the acoustic signals collected for each state (ST120).

Subsequently, the plant fault diagnosis apparatus 100 extracts a frequency equal to or greater than a preset reference magnitude value from the frequency spectrum for each condition, thereby setting an input frequency domain for each diagnosis condition including the extracted frequency (ST130).

The plant fault diagnosis apparatus 100 extracts the minimum and maximum values, that is, the minimum and maximum frequencies, as fuzzy parameters in the input frequency domain (ST140).

In addition, the plant defect diagnosis apparatus 100 generates first to third membership functions for each diagnosis state by applying the fuzzy parameter for each diagnosis state extracted in step ST140 to a preset fuzzy function (ST150).

4 and 5 are diagrams for explaining in more detail a process of generating a membership function.

4 is an FFT obtained by converting an acoustic signal (A) in different diagnostic states, that is, a safety, warning, or danger state, and an acoustic signal (A) in each state by the frequency converter 120 to FFT (fast Fourier transform). It is a diagram illustrating an input frequency determined from the spectrum (B) and the FFT spectrum (B) for each state.

As shown in FIG. 4, all frequencies in which a preset reference value, for example, a magnitude value (Y-axis) of "0.5" or higher in the FFT spectrum (B), may be extracted as an input frequency.

In Fig. 4, the input value of the low frequency band (1KHz or less) is extracted in the stable state for the plant to be diagnosed, the input value of the middle frequency band (4KHz region) is extracted in the warning state, and the input of the high frequency band is in the dangerous state. It can be seen that the value (area exceeding 5KHz) is extracted.

5 is a diagram for explaining a fuzzy function for generating a membership function for diagnosing defects in plant facilities. The fault diagnosis unit 130 diagnoses the fault condition based on Mamdani fuzzy inference about the input frequency. At this time, the defect status for the plant equipment is classified into a safety status, a warning status, and a dangerous status.

Referring to FIG. 5, it is possible to apply a fuzzy function in which frequencies of input frequency domains for each state (safe, warning, dangerous) belong to different regions.

Equation 1 below is a fuzzy function for determining a defect state for a plant facility.

In Equation 1, a is a minimum value of a frequency band corresponding to each diagnosis area, b is a maximum value of a frequency band corresponding to each diagnosis area, and x is an input frequency (input value) to be diagnosed.

As shown in FIG. 5, the fuzzy function allows frequencies D of different input frequency bands to belong to different diagnostic areas E, and output them as belonging values that can express the degree of belonging.

That is, the fault diagnosis unit 130 determines whether the input frequency is in a stable state by using the first membership function in which the minimum and maximum values corresponding to the safe region for one frequency are applied to Equation 1, and is stable. If it is not in a warning state, the minimum value and the maximum value corresponding to the warning region are determined by using the second membership function applied to Equation 1 above, and if it is not in the warning state, the minimum value corresponding to the danger region is determined. Using the third membership function applied to Equation (1) and the maximum value, it is determined whether the input frequency is in a dangerous state.

In addition, the fault diagnosis unit 130 applies all frequencies determined as fuzzy input values to Equation 1 to determine whether there is a stable, warning, or dangerous state.

Meanwhile, FIG. 6 is a diagram for explaining a process of determining the current state of the plant shown in FIG. 2, and is a diagram illustrating a result of applying a diagnostic function to a warning and a dangerous state in the defect diagnosis unit 130.

In FIG. 6, (X) is a diagram illustrating the diagnosis result for each input frequency (1-6) corresponding to the warning state, and (Y) is the diagnosis result for the input frequency (1-6) corresponding to the dangerous state. It is a drawing which illustrates.

1 to 6 applied to (X) and (Y) are the same input frequency, L1 is the reference value of the diagnostic area corresponding to the warning state, and L2 is the reference value of the diagnostic area corresponding to the dangerous state. L1 and L2 are the center value of the warning area and the center value of the danger area in FIG. 5E.

That is, as a result of applying the input frequencies 1 to 6 to the second membership function corresponding to the warning state, the third input frequency has a high membership value in the warning region. Accordingly, the warning state diagnosis result WR is the third input frequency. It may be determined as a membership value corresponding to.

In addition, as a result of applying the input frequencies 1 to 6 to the third membership function corresponding to the dangerous state, the sixth input frequency has a high membership value in the dangerous region. Accordingly, the dangerous state diagnosis result DR is the sixth input frequency. It may be determined as a membership value corresponding to.

Based on the result of this belonging value, the defect diagnosis unit 130 finally determines the diagnosis result to a dangerous state having a higher belonging value among the warning state diagnosis result and the dangerous state diagnosis result for the acoustic signal corresponding to the 1 to 6 input frequencies. Decide.

100: plant fault diagnosis device, 110: sound collection unit,
120: signal conversion unit, 130: defect diagnosis unit,
140: result output unit, 150: data memory,
1: Plant.

Claims (5)

A first step of generating and storing first to third membership functions respectively corresponding to sound signals for each diagnosis state of safety, warning, and danger in the plant defect diagnosis device; and
A second step of collecting an acoustic signal generated when the plant to be diagnosed is driven by the plant defect diagnosis device,
A third step of obtaining a frequency spectrum by FFT-converting the acoustic signal collected by the plant fault diagnosis apparatus, and determining a frequency having a preset magnitude value or more in the frequency spectrum as an input frequency, and
In the plant fault diagnosis apparatus, the input frequency is applied to the first to third membership functions, respectively, to calculate the first to third membership values, and to determine the diagnosis state corresponding to the membership function calculated for the maximum membership value as the plant diagnosis state. Consisting of including a fourth step,
In the first step, the first to third membership functions are functions for outputting the degree to which the input frequency belongs in the diagnosis area of the corresponding membership function as a membership value, and is generated by the following equation. How to diagnose defects in plant equipment.

In the above equation, a and b are the minimum and maximum frequencies corresponding to the acoustic signal region for each diagnosis state, and x is the input frequency.
delete The method of claim 1,
The first step is a step of collecting sound signals for each diagnostic state in a plant fault diagnosis apparatus,
FFT conversion of the collected acoustic signals for each diagnosis state to obtain a frequency spectrum for each diagnosis state, and setting a frequency domain belonging to each diagnosis state in each frequency spectrum,
Extracting the minimum and maximum values of the frequency domain belonging to each diagnosis status as a fuzzy parameter, and
And generating first to third membership functions by applying fuzzy parameters for each diagnosis state to the above equations, respectively.
The method of claim 1,
In the third step, when the number of input frequencies extracted by the preset reference magnitude value exceeds the reference number range, the reference magnitude value is set high by a certain unit and the number of input frequencies is less than the reference number range. A method for diagnosing defects of a plant facility using an acoustic signal, characterized in that the reference magnitude value is set low in a predetermined unit so that the number of input frequencies satisfies the reference number range.
The method of claim 1,
In the fourth step, the plant fault diagnosis apparatus applies a plurality of input frequencies to the first to third membership functions, respectively, calculates frequency membership values for each input frequency for each membership function, and calculates the average of the frequency membership values for the corresponding membership. A method for diagnosing a defect of a plant facility using an acoustic signal, characterized in that it is determined as a belonging value for a function.

Images