JPH09196825A - Plant abnormality detection device - Google Patents

Abstract

(57) [Abstract] [Problem] To monitor a plant and to detect abnormalities, reduce false alarms, and enable stable abnormality detection. SOLUTION: A neuro device 5A that performs not only image processing by a camera 1 and an image processing device 3 but also acoustic processing by a microphone 2 and an acoustic processing device 4 and performs image processing to detect normality / abnormality as a result, acoustics. A neuro device 5 that performs processing and detects whether the result is normal or abnormal.
By adding B and the neurolearning device 7 to perform learning, the accuracy of abnormality detection is improved and stability is secured.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention installs an ITV (industrial television) camera and a microphone as a sound collecting means in various facilities (plants) on site to monitor the plant, and also uses a neuro learning function. The present invention relates to an abnormality detection device capable of detecting abnormality in a plant.

[0002]

2. Description of the Related Art Conventionally, there has been known one which monitors an oil leak, a steam leak, a generator slip ring spark or smoke in a power plant using an image processing technique (for example, Japanese Patent Laid-Open No. 6-266983). ), And further,
It is also known to perform processing of acoustic signals such as steam leak noise and power generator operating noise in a power generation plant together with the image processing technique as described above (for example, Japanese Patent Laid-Open No. H06-6 / 1994).
-266982 reference).

[0003]

However, the above problems have the following problems. (1) In the monitoring using the image processing technology, the detection rate of the abnormality is deteriorated due to the change of the illumination state, the contamination of the equipment, etc., which causes a false alarm. Also, in the case of processing acoustic signals,
False alarms due to disturbances other than sound related to the power plant become a problem. (2) When an abnormality is detected and notified to an observer or an inspector,
The normal / abnormal status varies depending on the amount of abnormality (for example, the amount of oil leakage). Moreover, there is a difference in the judgment level of normality and abnormality depending on the monitoring or inspector of each plant, and it is difficult to make a stable judgment. Therefore, an object of the present invention is to reduce false alarms and enable stable abnormality detection.

[0004]

In order to solve such a problem, according to the invention of claim 1, an image processing device for processing an image signal of a plant from an image pickup device, and a plant based on the processing result and the image signal are provided. A first neural network for judging normality / abnormality, an acoustic processing device for processing the acoustic signal of the plant from the sound collecting means, and a second neural network for judging normality / abnormality of the plant based on the processing result and the acoustic signal. A network and the first and second
Comprehensive judgment means for making a comprehensive judgment on the basis of the output from the neural network and learning means for learning the difference between the comprehensive judgment result and the actual situation of the plant. It is characterized in that it is given to the second neural network. According to the invention of claim 1, at least the image pickup device and the sound collecting means are mounted on a vehicle and can be configured to be movable between plants (invention of claim 2).

That is, by using the image processing technology and the sound processing technology together, and by utilizing the learning function by the neural network, the accuracy of abnormality detection is improved, and it is possible to cope with the actual changes in the on-site environment. . Moreover, the re-learning function by the neural network is used so that the number of repetitions of false alarms can be significantly reduced. Furthermore, it is possible to incorporate the know-how and the like of the monitoring or the experience of the inspector so that the reliability can be further improved.

[0006]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS FIG. 1 is a system configuration diagram showing a first embodiment of the present invention. In the figure, 1 is a television camera as an image pickup means, 2 is a microphone as a sound pickup means, 3 is an image processing device, 4 is an acoustic processing device, 5
Reference numerals A and 5B are neural networks (also called neuro devices) well known as one of artificial intelligence (AI), 6 is a comprehensive judgment device, 7 is a neuro learning device, and 8 is a monitoring or inspector. In FIG. 1, for example, oil leaks, steam leaks, or pulverized coal leaks in a thermal power plant are detected by image processing, and abnormal noise around the generator boiler and the mill is detected by acoustic processing.

That is, the video signal from the television camera 1 is given to the image processing device 3 and the neuro device 5A, and the acoustic signal from the microphone 2 is given to the acoustic processing device 4 and the neuro device 5B, respectively, and it is judged whether the signal is normal or abnormal. Is performed. That is, the image processing device 3 and the sound processing device 4 determine normality / abnormality, and when it is determined to be abnormal, the fact is notified to the neuro devices 5A and 5B, and the normality / abnormality is also determined here. . The outputs of the neuro devices 5A and 5B are input to the comprehensive judgment device 6 where a comprehensive judgment is made.

The monitoring or inspecting person 8 correlates the output from the comprehensive judgment device 6 with the situation of the site, and when it judges that it is normal but abnormal or vice versa, The effect is given to the neuro learning device 7, and the fact that the comprehensive judgment result has an error is incorporated into the neuro devices 5A and 5B as a learning result. Hereinafter, such learning operation is repeated to improve detection accuracy and reduce false alarms as much as possible.

A specific example will be described with reference to FIG. (1) About heavy oil / light oil leak To detect an oil leak, install an oil pan 10 (see FIGS. 2 (a) and 2 (b)) at the leak target location.
It detects heavy oil and light oil leaking to the. Since the light oil is transparent, an appropriate absorbent material is laid in the oil pan 10 to detect it from the state of soaking into the absorbent material. As the detection method, the image processing apparatus 3 uses a difference method, and the neuro apparatus 5A uses a grayscale change comparison method.

(1-1) Difference Method In the difference method, a normal image is stored as shown in FIG. 2A, and the difference from the input image at the time of determination as shown in FIG.
The determination is made as shown in (c) and compared with a reference value. (1-2) Gradation Change Comparison Method In the same manner as above, the changed area of the image is obtained (see FIG. 2B), and this grayscale image is input to the neuro device 5A to determine abnormality. As the neuro device 5A, here, for example, as shown in FIG. 2D, a device whose network structure is a three-layer structure of an input layer, an intermediate layer, and an output layer is used. Further, the neuro device 5A is provided with a learning function, and learning is performed by the output from the neuro learning device 7.

(2) Abnormality of Steam Leakage A concrete example is shown in FIG. 3, which is a method for capturing an image of a large change in shape such as steam or smoke, or a fluctuating object. (2-1) Reference image update method Images of several frames immediately before and after the determination (± Δt) (see FIG. 3).
(B) and (c) are taken in, and the difference from the image at the time of determination as shown in FIG. Get like (e). Then, the logical sum (O
R) or an exclusive OR (XOR) or other logical processing is performed to obtain an image as shown in FIG. 3F, and this is compared with a reference image to make a determination. (2-2) Gradation Change Comparison Method A change area of an image obtained by the same method as described above, for example, a logical operation image as shown in FIG. 3G is converted into a neuro device as shown in FIG. Input) and judge the abnormality.

(3) Leakage of pulverized coal (detection of blown state) Here, a method of catching a change in contrast is used. A specific example is shown in FIG. (3-1) Contrast Change Extraction Method The density histogram of the image as shown in FIG. 4 (a) is linearly expanded to obtain a contrast enhanced image as shown in FIG. 4 (b). The emphasized image is detected by comparing it with the reference image and the abnormality is determined. (3-2) Shade change comparison method A change area of an image obtained by the same method as described above, for example, an image as shown in FIG. 4C is changed to a neuro device as shown in FIG. 4D (see reference numeral 5A in FIG. 1). Input and judge abnormality.

(4) Leakage of pulverized coal (residual detection in the vicinity of equipment) Here, a method of catching a large change in shape or a fluctuating one is used. As shown in FIG. 5, there are the following two types. (4-1) Reference image updating method Images of several frames immediately before and after the determination (± Δt) (see FIG. 5).
(B) and (b) are taken in, and the difference from the image at the time of determination as shown in FIG. Get like (e). Thereafter, the difference images 1 and 2 are judged to be abnormal based on the logical sum image as shown in FIG. (4-2) Gradation Change Comparison Method A change area of an image obtained by the same method as described above, for example, a logical operation image as shown in FIG. 5 (g) is displayed on the neuro device of FIG. 5 (h) (see reference numeral 5A in FIG. 1). Input to and determine the abnormality.

Next, the sound processing method will be described. Regarding this, there is no clear algorithm for abnormality determination, so the following is performed here. Microphone 2 of FIG.
The neural network of FIG. 6B, which is capable of processing the waveform of FIG. 6A, which is obtained by frequency-analyzing the acoustic signal collected in 1. and extracting its characteristics without a clear algorithm ( It is recognized by the reference numeral 5B in FIG. 1), and the degree of “unusual” is detected. By repeating such processing, the field applicability is improved. Also, considering the combination of operating equipment, etc., a comparison display with the waveform in the case of “normal” is performed.

FIG. 7 is a schematic diagram showing a second embodiment of the present invention. In this system, a communication device 9 including a transmitter and a receiver in addition to the camera 1 and the microphone 2 is mounted on a vehicle 11 so that the monitoring areas 12A and 12B can be freely monitored. This is an example of detection. Needless to say, the various devices and equipment as shown in FIG. 1 may be mounted on the vehicle 11 and the communication device may be omitted.

[0016]

According to the present invention, the following effects can be expected. (1) Since the learning function by the neural network is used, it is possible to respond to actual changes in the on-site environment.
The accuracy of abnormality detection is improved. (2) The number of repeated false alarms can be greatly reduced by the re-learning function using the neural network.
In addition, it becomes possible to incorporate the know-how gained from monitoring or inspectors' experience, which further improves reliability.

[Brief description of drawings]

FIG. 1 is a system configuration diagram showing a first embodiment according to the present invention.

FIG. 2 is an explanatory diagram for explaining a specific example of heavy oil / light oil leak detection.

FIG. 3 is an explanatory diagram for explaining a specific example of vapor leak detection.

FIG. 4 is an explanatory diagram for explaining a specific example of pulverized coal leakage (blowing) detection.

FIG. 5 is an explanatory diagram for explaining a specific example of pulverized coal leakage (residual) detection.

FIG. 6 is an explanatory diagram illustrating an example of acoustic processing.

FIG. 7 is a system configuration diagram showing a second embodiment according to the present invention.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1 ... Tele camera, 2 ... Microphone, 3 ... Image processing device, 4 ... Sound processing device, 5A, 5B ... Neural network (neuro device), 6 ... Comprehensive judgment device, 7 ... Neuro learning device, 8 ... Monitoring or inspector, 9 ... communication device,
10 ... Oil pan, 11 ... Vehicle, 12A, 12B ... Monitoring area.

─────────────────────────────────────────────────── ─── Continuation of the front page (51) Int.Cl. ⁶ Identification code Internal reference number FI Technical display location G06T 1/00 G08B 25/00 510F G08B 25/00 510 G06F 15/62

Claims (2)

[Claims] 1. An image processing apparatus for processing an image signal of a plant from an image pickup apparatus, a first neural network for judging normality / abnormality of the plant based on the processing result and the image signal, and a plant for collecting sound. Sound processing apparatus for processing the sound signal of the above, a second neural network for judging normality / abnormality of the plant based on the processing result and the sound signal, and a comprehensive based on the outputs from the first and second neural networks. And a learning means for learning the difference between the comprehensive judgment result and the actual situation of the plant, and the learning result is given to the first and second neural networks. Anomaly detection device for plant. 2. The abnormality detecting device for a plant according to claim 1, wherein at least the image pickup device and the sound collecting means are mounted on a vehicle so as to be movable between the plants.