KR20230064416A - Apparatus and method for diagnosing fault of power system - Google Patents

Abstract

The present invention relates to a device and a method for diagnosing a fault of a power device. According to an embodiment of the present invention, the method comprises the steps of: obtaining data related to partial discharge of a power device through at least one sensor; separately predicting a probability value that the obtained data corresponds to known defect types using a plurality of neural network models; determining a final probability value for each of the defect types by integrating probability values separately predicted by the neural network models; calculating the reliability of a prediction result of the neural network models; confirming whether the calculated reliability is less than a specified threshold; and determining a defect type of the data as an unknown class when the calculated reliability is less than the threshold. Therefore, classification performance for a learned defect type can be maintained.

Description

Fault diagnosis apparatus and method of power equipment {APPARATUS AND METHOD FOR DIAGNOSING FAULT OF POWER SYSTEM}

The present invention relates to an apparatus and method for diagnosing a fault in a power device capable of diagnosing (eg discriminating or classifying) a partial discharge of an unknown (or unlearned) type (or class).

Power facilities such as power plants and substations include various power devices. Power devices include many components that operate at high voltages and/or high currents. When a high current flows through a power device, overheating may result in fire or equipment damage. Power equipment includes devices (e.g., electromechanical devices composed of high-voltage components such as circuit breakers and disconnectors) to quickly and timely prevent high currents from causing unpredictable damage. For example, the power device may include a gas insulated switchgear (GIS).

Meanwhile, a partial discharge occurs in a power device as a precursor to a failure. That is, when a partial discharge occurs in a power device, insulation breakdown may be reached, surface discharge may be caused, and finally, the power device may be damaged and operation may be stopped. Therefore, there is a need for a method for detecting and diagnosing partial discharge to prevent failure of power devices by diagnosing the occurrence of partial discharge at an early stage and taking appropriate measures. For example, methods for detecting and diagnosing various parts using ultra-high frequency (UHF), current induction, and acoustic emission are provided.

Recently, machine learning (or deep learning) techniques (e.g., artificial neural networks (ANNs), support vector machines (SVMs), convolutional neural networks (CNNs), deep neural networks (DNNs)) are partially based. Partial discharge defect diagnosis (e.g., defect type) that improves the accuracy of partial discharge failure type analysis and classification by analyzing and classifying the discharge failure type and reflecting the user's judgment results through machine learning (Machine Learning). Judgment) devices and methods are provided.

However, the conventional apparatus and method for diagnosing partial discharge accurately classify trained (or known) fault types, but fail to classify untrained (or unknown) fault types or erroneously classify them as trained fault types. There is a problem and/or discomfort with sorting. Accordingly, there is a demand for a defect diagnosis apparatus and method capable of distinguishing partial discharge of an untrained defect type.

Accordingly, an object of the present invention is to solve the above problems, and to provide an apparatus and method for diagnosing defects in power equipment capable of classifying the types of partial discharge into a trained type and an untrained type. It is to do.

To achieve the above object, a method for diagnosing a defect in a power device of the present invention includes acquiring data related to partial discharge of the power device through at least one sensor; predicting probability values corresponding to known defect types, respectively, using a plurality of neural network models; Determining a final probability value corresponding to each defect type by integrating probability values predicted by the plurality of neural network models; calculating reliability of prediction results of the plurality of neural network models; checking whether the calculated reliability is less than a specified threshold; and determining the defect type of the data as an unknown class when the calculated reliability is less than the threshold.

An apparatus for diagnosing a fault in a power device of the present invention includes a sensor module that obtains data related to a partial discharge of the power device; and predicting probability values corresponding to known defect types by using a plurality of neural network models, respectively, and integrating probability values respectively predicted by the plurality of neural network models to obtain a final probability value corresponding to each defect type. A probability value is determined, reliability of prediction results of the plurality of neural network models is calculated, it is determined whether the calculated reliability is less than a specified threshold, and if the calculated reliability is less than the threshold, the defect type of the data is unknown. It can include a fault diagnosis module that determines the type (unknown class).

As described above, the present invention can reduce overconfidence predictions by considering uncertainty estimates. That is, the present invention can prevent misclassification of a defect of a non-learned type as one of defects of a learned type. In addition, the present invention can maintain classification performance for defects of the learned type.

1 is a block diagram showing the configuration of a defect diagnosis apparatus according to an embodiment of the present invention.
2A to 2G are graphs illustrating patterns of partial discharge signals having various defect types according to an embodiment of the present invention.
3 is a diagram illustrating a distribution of partial discharge signals having various defect types according to an embodiment of the present invention.
4 is a diagram illustrating amplitudes of partial discharge signals having various defect types according to an embodiment of the present invention.
5 is a diagram showing the structure of a defect diagnosis module according to an embodiment of the present invention.
6 is a diagram illustrating a confusion matrix for classification performance of a defect diagnosis apparatus for which a threshold is not set.
7A and 7B are diagrams illustrating a confusion matrix for classification performance of a defect diagnosis apparatus for which a threshold is set according to an embodiment of the present invention.
8 is a diagram illustrating evaluation results of classification accuracy of a defect diagnosis apparatus according to an embodiment of the present invention.
9 is a diagram illustrating analysis results of sensitivity and precision of a defect diagnosis apparatus according to an embodiment of the present invention.
10 is a flowchart illustrating a method of determining a defect type of partial discharge according to an embodiment of the present invention.

Objects and effects of the present invention, and technical configurations for achieving them will become clear with reference to embodiments described later in detail in conjunction with the accompanying drawings. In describing the present invention, if it is determined that a detailed description of a known function or configuration may unnecessarily obscure the gist of the present invention, the detailed description will be omitted.

In addition, terms to be described later are terms defined in consideration of functions in the present invention, which may vary according to the intention or custom of a user or operator.

However, the present invention is not limited to the embodiments disclosed below and may be implemented in a variety of different forms. Only these embodiments are provided to complete the disclosure of the present invention and to fully inform those skilled in the art of the scope of the invention, and the present invention is defined by the scope of the claims. It only becomes. Therefore, the definition should be made based on the contents throughout this specification.

1 is a block diagram showing the configuration of a defect diagnosis apparatus according to an embodiment of the present invention, and FIGS. 2A to 2G show patterns of partial discharge signals having various defect types according to an embodiment of the present invention. 3 is a diagram showing the distribution of partial discharge signals having various defect types according to an embodiment of the present invention, and FIG. 4 is a diagram showing partial discharge signals having various defect types according to an embodiment of the present invention. A diagram showing the amplitude, and FIG. 5 is a diagram showing the structure of a defect diagnosis module according to an embodiment of the present invention.

1 to 5, the fault diagnosis apparatus 100 according to an embodiment of the present invention is included in a power facility providing high power such as a power plant or a substation, and monitors power devices constituting the power facility, Partial discharge (PD) faults can be diagnosed (ie detected and typed). To this end, the fault diagnosis apparatus 100 may include a power cutoff module 110 and a fault diagnosis module 130 .

The power cutoff module 110 is a protection device that prevents damage to power devices from abnormalities such as facility failure, lightning, short circuit, and electric leakage. For example, the power cutoff module 110 may be a gas insulated switchgear (GIS). This is just one example, and the power cutoff module 110 may include various devices capable of cutting off power.

Various types of partial discharge may occur in the power cutoff module 110 as a precursor of failure. For example, the power cut-off module 110, as shown in FIGS. 2A to 2G , crack-type partial discharge (FIG. 2A), floating-type partial discharge (FIG. 2B), free particle type Partial discharge (FIG. 2c), void type partial discharge (FIG. 2d), POC (protrusion on conductor) type partial discharge (FIG. 2e), POS (particle on spacer) type partial discharge (FIG. 2f) or A protrusion on enclosure (POE) type of partial discharge (Fig. 2g) may occur. This is only an example, other types of partial discharges may occur.

Partial discharge may cause deterioration of an insulator and ultimately cause failure/damage of the power cutoff module 110 . To prevent failure/damage by recognizing partial discharge in advance, the power cutoff module 110 may include a sensor module 120 for detecting partial discharge. The sensor module 120 is an ultra-high frequency (UHF) sensor that detects a partial discharge signal (eg, a signal operating in an ultra-high frequency band (0.5 to 1.5 GHz) and having a sensitivity level of -14.5 dBm in 5 power cycles) can This is just an example, and the sensor module 120 may include various sensors for detecting the partial discharge signal. Meanwhile, the sensor module 120 may be located outside the power cutoff module 110 .

The defect diagnosis module 130 may receive the sensor data obtained by the sensor module 120 and analyze the sensor data to determine whether there is a partial discharge defect and the type of defect. For example, the fault diagnosis module 130 captures the maximum value of the partial discharge signal using a peak detector and generates a phase resolution partial discharge (PRPD) pattern using N samples in each power cycle. can be measured

Meanwhile, partial discharge signals of various defect types may have similar characteristics (eg, t-distributed stochastic neighbor embedding (t-SNE) characteristics) and amplitudes. For example, as shown in FIG. 3, the partial discharge signals of various defect types may have a short distance between the partial discharge signals of various defect types, and as shown in FIG. 4, the average amplitude level (horizontal axis ) and the maximum amplitude level (vertical axis) may be similar. As described above, due to similarity in characteristics and amplitudes of partial discharge signals of various types of defects, it is difficult for the defect diagnosis module 130 to accurately classify the type of partial discharge using the partial discharge signal measured through the sensor module 120. . For example, the fault diagnosis module 130 may misclassify a partial discharge of an untrained (unknown) type as a partial discharge of a trained type. A detailed description thereof will be described later with reference to FIG. 6 .

The defect diagnosis module 130 according to an embodiment of the present invention may classify the detected partial discharge type (or class) based on uncertainty estimation. For example, the fault diagnosis module 130 may classify the partial discharge into one of known classes or an unknown class.

The defect diagnosis module 130 may have a deep ensemble structure (deep ensemble network) including a plurality of neural network models (eg, CNN models). For example, the defect diagnosis module 130 may include an input unit 131, a plurality of CNN models 133, and an ensemble output unit 135, as shown in FIG. 5 .

The input unit 131 may transmit the sequential phase-resolved partial discharge signal X to a plurality of neural network models 133 . Here, the phase-resolved partial discharge signal (X) is equal to <Equation 1> below.

X = M x N ............ <Equation 1>

In <Equation 1>, M is the number of power cycles and N is the number of phase angles.

The plurality of neural network models 133 may classify the defect type of the partial discharge (calculate a probability value that the input partial discharge signal corresponds to each type) based on the phase-resolved partial discharge signal X. Parameters of each neural network model included in the plurality of neural network models 133 may be optimized through a training phase based on various partial discharge data. Each neural network model has two convolution layers (133a) with a kernel size of 3 x 3, two max pooling layers (133b) with a kernel size of 3 x 3, and two dropout layers (133c) for normalization. , a fully connected layer 133d having 128 nodes, a fully connected layer 133e having 64 nodes, and an output layer 133f outputting the probability of each class. In this case, the output layer 133 may calculate a relative probability value of each class (probability value that the input partial discharge signal corresponds to each type) through the softmax function of Equation 2 below.

……………<식 2>

… … … … … <Equation 2>

In <Equation 2>, k is an index of a class, and may have values of k = 1, 2, ..., K. K is the total number of classes. zk is an input of an activation function (convolutional layer 133a).

The ensemble output unit 135 may calculate an average value of output values of a plurality of neural network models 133 as shown in Equation 3 below.

……………<식 3>

… … … … … <Equation 3>

는 h번째 뉴럴 네트워크 모델의 출력 값이고, H는 뉴럴 네트워크 모델의 총 수이다.In <Equation 3>,

is the output value of the h-th neural network model, and H is the total number of neural network models.

)에 기초하여 부분방전의 클래스를 결정할 수 있다. 신뢰도(

)는 아래의 <식 4>와 같이 정의될 수 있다. 즉, 신뢰도(

)는 각 클래스에 대한 확률 값들 중 최대값으로 결정될 수 있다.The defect diagnosis module 130 determines the reliability of the output value of the ensemble output unit 135 (

), the class of partial discharge can be determined based on Reliability (

) can be defined as in <Equation 4> below. That is, reliability (

) may be determined as the maximum value among probability values for each class.

..............<식 4>

..............<Formula 4>

Meanwhile, when the class is determined only by reliability without a threshold, the defect diagnosis module 130 may misclassify an unknown class (or an untrained class) as one of the known classes despite low similarity. A description thereof will be described in detail with reference to FIG. 6 .

)와 지정된 임계치(Thc)를 비교하여 부분방전의 클래스를 분류할 수 있다. 예를 들어, 결함 진단 모듈(130)은 신뢰도(

)가 지정된 임계치보다 작으면, 부분방전을 알려지지 않은 클래스로 분류하고, 신뢰도(

)가 지정된 임계치(Thc) 이상이면, 알려진 클래스들 중 하나로 분류할 수 있다. 상기 임계치는 현실 세계에서 획득된 부분방전 데이터(예: 위상분해 부분방전 데이터)를 이용하여 결정될 수 있다.In order to solve the misclassification problem, the defect diagnosis module 130 of the present invention has reliability (

) and the designated threshold value (Thc), the partial discharge class can be classified. For example, the fault diagnosis module 130 determines the reliability (

) is less than the specified threshold, classify the partial discharge into an unknown class, and determine the reliability (

) is equal to or greater than the specified threshold (Thc), it can be classified into one of the known classes. The threshold may be determined using partial discharge data (eg, phase-resolved partial discharge data) acquired in the real world.

Meanwhile, the fault diagnosis module 130 may be a part of a main processor (not shown) that controls the power device, a separate processor, or a software module.

6 is a diagram illustrating a confusion matrix for classification performance of a defect diagnosis apparatus for which a threshold is not set.

Prior to the detailed description, the defect diagnosis module learned only partial discharge data of crack, floating, free particle, and void types among the seven types of partial discharges shown in FIGS. 2A to 2G, and POC, POS, and POE types. were unlearned about them. That is, the Crack, Floating, Free Particle, and Void types are known types of partial discharges, and the POC, POS, and POE types are unknown types of partial discharges.

Referring to FIG. 6 , the fault diagnosis module for which no threshold is set detects a partial discharge of an unknown type (POC) as a known type (free particle ) is misclassified as a partial discharge. As another example, the fault diagnosis module for which no threshold is set detects a partial discharge of an unknown type (POE) with a probability of 75%, as shown in the figure 620 of FIG. 6, of a crack type among known types. It is misclassified as a discharge, and it is misclassified as a partial discharge of free particle type with a probability of 25%. As another example, the fault diagnosis module for which no threshold is set detects a partial discharge of an unknown type (POS) with a probability of 28%, as shown in the figure 630 of FIG. 6, of a crack type among known types. It is misclassified as a partial discharge and misclassified as a free particle type partial discharge with a probability of 72%.

7A and 7B are diagrams illustrating a confusion matrix for classification performance of a defect diagnosis apparatus for which a threshold is set according to an embodiment of the present invention.

Prior to the detailed description, partial discharge type classification was performed while setting the threshold of the defect diagnosis module to 95% and changing the number of neural network models to 1, 5, or 10.

Referring to FIGS. 7A to 7C , it can be seen that the classification accuracy of the defect diagnosis apparatus varies according to the number of neural network models. First, when the number of neural network models is 1, the defect diagnosis module detects a partial discharge of an unknown type (POC) as a known type (free particle) with a probability of 100%, as shown in the diagram of identification code 710 of FIG. 7A. is misclassified as a partial discharge of As another example, the fault diagnosis module in which the threshold is set, as shown in the diagram of identification code 720 of FIG. 7A, partial discharge of an unknown type (POE) is classified as a partial discharge of a crack type among known types with a probability of 75%. misclassify, and with a probability of 25% misclassify as partial discharge of free particle type. As another example, the fault diagnosis module in which the threshold is set, as shown in the diagram of identification code 730 of FIG. 7A, partial discharge of an unknown type (POS) with a probability of 20% is a partial discharge of a crack type among known types. , misclassified as a partial discharge of free particle type with a probability of 72%, and classified as a partial discharge of an unknown type with a probability of 8%.

Next, when the number of neural network models is 5, the defect diagnosis module detects a partial discharge of an unknown type (POC) as a known type (free particle ), and classified as an unknown type of partial discharge with a probability of 64%. As another example, the fault diagnosis module in which the threshold is set, as shown in the drawing of identification code 750 in FIG. 7B , partial discharge of an unknown type (POE) is classified as a partial discharge of a crack type among known types with a probability of 70%. misclassify, with a probability of 30 % classified as a partial discharge of unknown type. As another example, the fault diagnosis module for which the threshold is set misclassifies a partial discharge of an unknown type (POS) as a partial discharge of a free particle type with a probability of 20%, as shown in the diagram of identification code 760 of FIG. 7B . and classified as a partial discharge of unknown type with a probability of 80 %.

Finally, when the number of neural network models is 10, the defect diagnosis module detects a partial discharge of an unknown type (POC) with a probability of 100%, as shown in the drawing of identification code 770 in FIG. 7c. classified as discharge. As another example, the fault diagnosis module for which the threshold is set classifies a partial discharge of an unknown type (POE) as a partial discharge of an unknown type with a probability of 100%, as shown in the diagram of identification code 780 in FIG. 7C . As another example, the fault diagnosis module for which the threshold is set classifies a partial discharge of an unknown type (POS) as a partial discharge of an unknown type with a probability of 100%, as shown in the drawing of identification code 790 in FIG. 7C . .

Meanwhile, referring to FIGS. 7A to 7C , it can be seen that classification results for known types of partial discharges (crack, floating, free particle, and void) are maintained even if the number of neural network models is changed. That is, by properly setting the threshold and the number of neural networks, it is possible to prevent partial discharges of unknown types from being misclassified as partial discharges of known types while maintaining classification performance for partial discharges of known types.

Meanwhile, FIGS. 7A to 7C are only examples, and the number of neural network models may be changed according to characteristics of input data and classification targets.

8 is a diagram illustrating evaluation results of classification accuracy of a defect diagnosis apparatus according to an embodiment of the present invention.

Prior to the detailed description, the classification accuracy of the fault diagnosis device is obtained by varying the threshold value (e.g. 0, 0.75, 0.85, 0.95) and the number of neural networks (e.g. 1, 5, 10), and the learned class (known type) and Evaluated for partial discharge of learned class (unknown type).

Referring to FIG. 8, the fault diagnosis apparatus maintains classification accuracy (at least 95.52%) for partial discharges of a learned class (known type) and prevents misclassification of partial discharges of a non-learned class (unknown type). (e.g., if the threshold is 0.95 and the number of neural networks is 10, it can be prevented by 100%).

9 is a diagram illustrating analysis results of sensitivity and precision of a defect diagnosis apparatus according to an embodiment of the present invention.

Prior to the detailed description, the sensitivity and precision of the fault diagnosis device were determined by setting the threshold to 0.95 and changing the number of neural network models (e.g., 1, 5, 10) to learn class (known type) and non-learned class. (of unknown type) was evaluated for partial discharge.

Sensitivity is measured by <Equation 5> below, and precision is measured by <Equation 6>.

……………… <식 5>

… … … … … … <Equation 5>

………………… <식 6>

… … … … … … … <Equation 6>

In <Equation 5> and <Equation 6>, TP (True Positive) means that the type of failure is accurately classified, FP (False Positive) means that it is classified as a failure when it is not an actual failure, and FN (False Negative ) means that the failure is classified into different failure types.

Referring to FIG. 9 , it can be seen that the sensitivity for the learned (or known) class (or type) remains the same even if the number of neural network models is changed. In contrast, it can be seen that the sensitivity for unlearned (or unknown) classes (or types) increases as the number of neural network models increases. In particular, when the number of neural network models is 10, it can be seen that the sensitivity for the non-learned class increases to 100%.

Meanwhile, it can be seen that the precision of the learned class is 100% when the number of neural network models is 10. In contrast, it can be seen that the precision for unlearned (or unknown) classes (or types) is over 90%. This is because, as the threshold value is set to 0.95, even a known class can be classified (predicted) into an unknown class when the reliability is less than 0.95.

10 is a flowchart illustrating a method of determining a defect type of partial discharge according to an embodiment of the present invention.

Referring to FIG. 10 , in step 1001, the defect diagnosis apparatus according to an embodiment of the present invention may obtain partial discharge data through a sensor (eg, the sensor module 120). For example, the defect diagnosis apparatus may obtain a partial discharge signal through an ultrasonic sensor.

In step 1003, the defect diagnosis apparatus may predict a probability value corresponding to each defect type with respect to the acquired data using a plurality of neural network models (eg, a plurality of neural network models 133). For example, each neural network model may predict a probability value corresponding to previously learned (or known) types of data, as shown in <Equation 2>. In other words, when a plurality of neural network models are trained for four defect types, each neural network model has a first probability that the data corresponds to the first type, a second probability that the data corresponds to the second type, and a third probability that the data corresponds to the second type. A third probability value corresponding to the type and a fourth probability value corresponding to the fourth type may be predicted. At this time, assuming that the combined diagnosis device includes 10 neural network models, a first probability value corresponding to the first type, a second probability value corresponding to the second type, and a third probability corresponding to the third type value and the fourth probability value corresponding to the fourth type are 10, respectively.

In step 1005, the defect diagnosis apparatus may determine a final probability value by integrating probability values for each type predicted by a plurality of neural network models. For example, as shown in <Equation 3>, the defect diagnosis apparatus may determine an average value of probability values for each type as the final probability value. In other words, the defect diagnosis apparatus includes 10 neural network models, and when the neural network models are trained for 4 types of defects, the average value of the 10 first probability values is determined as the final probability value corresponding to the first type. And, the average value of the 10 second probability values is determined as the final probability value corresponding to the second type, the average value of the 10 third probability values is determined as the final probability value corresponding to the third type, and the 10 An average value of the fourth probability values may be determined as a final probability value corresponding to the fourth type.

In step 1007, the defect diagnosis apparatus may calculate reliability of a prediction result using a plurality of neural network models. For example, the defect diagnosis apparatus may determine reliability as the largest value among final probability values, as shown in <Equation 4>.

In step 1009, the defect diagnosis apparatus may check whether the reliability is less than a specified threshold value (eg, 95%). The threshold may be determined based on partial discharge data collected in the real world. As a result of the confirmation in step 1009, if the reliability is less than the specified threshold, the defect diagnosis apparatus may determine the defect type of the data as an unknown defect type (or an unlearned defect type) in step 1011. On the other hand, if the reliability is equal to or greater than the specified threshold value, the defect diagnosis apparatus may determine the defect type of the data as one of known defect types (a defect type having the highest final probability value) based on the reliability in step 1013. . For example, the final probability value corresponding to the first type is 96%, the final probability value corresponding to the second type is 4%, the final probability value corresponding to the third type is 0%, and the final probability value corresponding to the fourth type is When the corresponding final probability value is 0%, the defect diagnosis apparatus may determine the defect type of the data as the first type.

Although the above has been described with reference to the illustrated embodiments of the present invention, these are only examples, and those skilled in the art to which the present invention belongs can variously It will be apparent that other embodiments that are variations, modifications and equivalents are possible. Therefore, the true technical protection scope of the present invention should be determined by the technical spirit of the appended claims.

100: fault diagnosis device
110: power blocking module 120: sensor module
130: fault diagnosis module

Claims (10)

A method for diagnosing a fault in a power device, comprising:
obtaining data related to partial discharge of the power device through at least one sensor;
predicting probability values corresponding to known defect types, respectively, using a plurality of neural network models;
Determining a final probability value corresponding to each defect type by integrating probability values predicted by the plurality of neural network models;
calculating reliability of prediction results of the plurality of neural network models;
checking whether the calculated reliability is less than a specified threshold; and
and determining a defect type of the data as an unknown class when the calculated reliability is less than the threshold.
According to claim 1,
and determining a defect type of the data as a defect type corresponding to the reliability among known defect types when the calculated reliability is greater than or equal to the threshold.
According to claim 1,
The step of determining the final probability value is
And determining an average value of probability values for each type as the final probability value.
According to claim 3,
The step of calculating the reliability is
And determining a maximum value among average values for each type as the reliability.
According to claim 1,
The plurality of neural network models are
Based on partial discharge data obtained in the real world, each method is trained separately and has different optimal parameters.
In the fault diagnosis device of power equipment,
a sensor module acquiring data related to partial discharge of the power device; and
Using a plurality of neural network models, probability values corresponding to known defect types are each predicted, and probability values respectively predicted by the plurality of neural network models are integrated to obtain a final probability value corresponding to each defect type. Determines, calculates the reliability of the predicted results of the plurality of neural network models, determines whether the calculated reliability is less than a specified threshold, and if the calculated reliability is less than the threshold, determines the defect type of the data as an unknown type. A device characterized in that it includes a fault diagnosis module that is determined as (unknown class).
According to claim 6,
The fault diagnosis module
When the calculated reliability is greater than or equal to the threshold value, a defect type of the data is determined as a defect type corresponding to the reliability among known defect types.
According to claim 6,
The fault diagnosis module
An apparatus characterized by determining an average value of probability values for each type as the final probability value.
According to claim 8,
The fault diagnosis module
An apparatus characterized by determining a maximum value among average values for each type as the reliability.
According to claim 6,
The sensor module
A device characterized in that it is an ultrasonic sensor.

Images