JP2022026110A - Abnormality assessment system and abnormality assessment method - Google Patents

Abstract

To provide an abnormality assessment system and an abnormality assessment method, enabling abnormality to be highly accurately assessed.SOLUTION: An abnormality assessment system 1 for treatment equipment 100 comprising a reaction vessel 2 and an air blower 3 is provided . The abnormality assessment system 1 comprises: an image data storage unit 8 preserving image data obtained by photographing the reaction vessel 2; an image processing unit 9 calculating the area of bubbles in the image data; a bubble area storage unit 10 preserving the area of bubbles; an air blown amount data storage unit 11 preserving the air blown amount of the air blower 3; a regression model creation unit 12 creating a regression model on the basis of the area of bubbles and the air blown amount; an air blown amount estimation unit 13 estimating the air blown amount using a newly acquired area of bubble and the regression model; and an abnormality assessment unit 14 assessing abnormality on the basis of the estimated air blown amount and the actual air blown amount.SELECTED DRAWING: Figure 1

Description

The present invention relates to an abnormality determination system and an abnormality determination method.

Conventionally, in order to treat sewage, sewage, etc., sewage is introduced into a reaction tank, and air is supplied (diffused) into the sewage by a blower. By dissipating air in the reaction tank, the purification action by microorganisms is maintained and promoted to purify sewage. The purified treated water is sent to the next process or discharged as it is. If an abnormality occurs in the air supply state during the treatment by this air diffuser, the purification function of the reaction tank is deteriorated or impaired. In that case, the sewage that is incompletely treated is sent to the next process and finally discharged to the outside, which causes environmental pollution.

Patent Document 1 discloses an abnormality detection device for detecting an abnormal state as described above. In this document, a barometric pressure sensor is connected to the air pressure pipe that supplies air from the blower to the air diffuser in the septic tank, and the air pressure in the air pressure pipe is observed. A device that outputs a barometric pressure abnormality signal when it is high or low is described.

Japanese Unexamined Patent Publication No. 2005-169310

In Patent Document 1, since the only index for detecting an abnormality is the air pressure in the air supply pipe, an abnormality occurs in the air pressure sensor that monitors the air pressure, or a failure occurs in the path of the air supply pipe connected to the air pressure sensor. In that case, it may not be possible to correctly grasp the actual operating condition of the septic tank.

The present invention has been made in view of the above-mentioned circumstances, and an object of the present invention is to provide an abnormality determination system and an abnormality determination method capable of performing abnormality determination with high accuracy.

The present invention employs the following means in order to solve the above problems.
That is, the abnormality determination system of the present invention is an abnormality determination system of a processing device provided with a reaction tank and a blower, and has an image data storage unit for storing image data obtained by photographing the reaction tank and bubbles of image data in the previous period. An image processing unit that calculates the area, a foam area storage unit that stores the area of the foam, an air volume data storage unit that stores the air volume of the blower, and a regression based on the bubble area and the air volume. An abnormality is generated based on the regression model creation unit that creates a model, the air flow estimation unit that estimates the air flow amount from the newly acquired bubble area and the regression model, and the estimated air flow amount and the actual air flow amount. It is provided with an abnormality determination unit for determination.
According to such an invention, a regression model is created from image data and airflow data, and an abnormality is determined based on the airflow amount estimated from the regression model and the actual airflow amount. It is possible to make a highly accurate abnormality determination based on a plurality of viewpoints.

In one aspect of the present invention, a water quality data storage unit for storing water quality data of the reaction tank is provided, and the regression model creation unit creates a regression model based on the water quality data, the area of the bubbles, and the amount of air blown. The blower amount estimation unit is created and estimates the blower amount from the newly acquired bubble area and water quality data and the regression model.
According to such a configuration, since the regression model is created based on the water quality data in addition to the area of the bubbles, it is possible to estimate the air flow amount more accurately, and the accuracy of the abnormality determination is improved.

The present invention is an abnormality determination system of a processing device provided with a reaction tank and a blower, and is an image data storage unit for storing image data obtained by photographing the reaction tank, and a blower amount data storage for storing the blower amount of the blower. A newly acquired unit, a water quality data storage unit that stores the water quality data of the reaction tank, and a deep learning model creation unit that creates a deep learning model based on the water quality data, the image data, and the air volume. It is provided with an airflow amount estimation unit that estimates the airflow amount from the image data, water quality data, and the deep learning model, and an abnormality determination unit that determines an abnormality based on the estimated airflow amount and the actual airflow amount.
According to such an invention, since the deep learning model is created based on the water quality data, the image data, and the air flow amount data, the procedure of data analysis can be simplified. Since the deep learning model automatically optimizes the parameters of the model, it is possible to estimate the amount of air blown with high accuracy and improve the accuracy of abnormality determination.

In one aspect of the invention, the image processing unit calculates the area of the foam by adaptive binarization.
According to such a configuration, the area of bubbles can be calculated accurately, and the accuracy of abnormality determination can be improved.

The method for determining an abnormality of the present invention is an abnormality determination method for a processing device provided with a reaction tank and a blower, in which the reaction tank is photographed, the area of bubbles in the image data taken in the previous period is calculated, and the blower is used. To measure the amount of air blown, create a regression model based on the area of the bubbles and the amount of air blown, estimate the amount of air blown from the area of bubbles in the newly captured image data and the regression model, It includes determining an abnormality based on the estimated air flow amount and the actual air flow amount.
According to such an invention, a regression model is created from image data, blast volume data, and water quality data, and an abnormality is determined based on the blast volume estimated from the regression model and the actual blast volume. It is possible to perform highly accurate abnormality determination based not only on the air volume but also on a plurality of viewpoints.

According to the present invention, it is possible to provide an abnormality determination system and an abnormality determination method capable of performing abnormality determination with high accuracy.

It is a schematic diagram which shows the whole structure including the abnormality determination system which concerns on embodiment of this invention. It is a functional block diagram which shows the structure of the abnormality determination system which concerns on embodiment of this invention. It is an image diagram for demonstrating the image conversion of the image processing part which concerns on embodiment of this invention. It is a graph for demonstrating the conversion of the recorded data which concerns on embodiment of this invention. It is a functional block diagram which shows the structure of the abnormality determination system which concerns on embodiment of this invention. It is a functional block diagram which shows the structure of the abnormality determination system which concerns on embodiment of this invention. It is a schematic diagram for demonstrating an example of the deep learning model which concerns on embodiment of this invention.

(First Embodiment)
Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings.
FIG. 1 is a schematic view showing an overall configuration including an abnormality determination system according to the present embodiment, and shows an abnormality determination system 1 of a processing device 100 provided with a reaction tank 2 and a blower 3. In addition to the reaction tank 2 and the blower 3, the processing device 100 includes an air diffuser 4 that diffuses air into the reaction tank 2, a valve 5 that opens and closes the flow of air from the blower 3 to the air diffuser 4, and a blower 3. It is equipped with an air volume measuring device 6 for measuring the amount of air blown to the air device 4, and a camera 7 for photographing the water surface of the reaction tank. The amount of air blown detected by the airflow amount measuring device 6 and the image data taken by the camera 7 are sequentially transmitted to the abnormality determination system 1.

The abnormality determination system 1 includes an image data storage unit 8, an image processing unit 9, a bubble area storage unit 10, an air volume data storage unit 11, a regression model creation unit 12, an air flow amount estimation unit 13, and an abnormality determination unit 14. .. The abnormality determination system 1 is an information processing device such as a personal computer. The image data storage unit 8, the air flow amount data storage unit 11, and the bubble area storage unit 10 are realized by storage devices such as a semiconductor memory and a magnetic disk provided inside and outside the information processing device. Further, the image processing unit 9, the regression model creation unit 12, the air flow amount estimation unit 13, and the abnormality determination unit 14 may be software or programs executed by the CPU or GPU in the information processing device.

As shown in FIG. 1, the sewage flowing into the reaction tank 2 is biologically treated and purified by microorganisms. At this time, microorganisms proliferate as they decompose organic matter, which forms activated sludge. When decomposing organic matter with aerobic microorganisms, it is necessary to send air to supply oxygen to the microorganisms in activated sludge. A plurality of air diffusers 4 are provided in the lower part of the reaction tank 2 so that air is supplied from the blower 3. An air volume adjusting valve 5 is provided in the pipe leading to the air diffuser 4, and the air volume is adjusted by opening and closing the valve 5. By adjusting the amount of air blown, the DO (Dissolved Oxygen) value, which is the amount of dissolved oxygen in the reaction tank 2, is adjusted, and the progress of biological treatment is adjusted. The blast amount is constantly acquired by the blast amount measuring device 6.

FIG. 2 is a functional block diagram showing the configuration of the abnormality determination system 1 in the present embodiment in more detail. The abnormality determination system 1 further includes a regression model parameter storage unit 12a that stores the parameters of the regression model created by the regression model creation unit 12.

As shown in FIG. 2, the blast amount measured by the blast amount measuring device 6 of FIG. 1 is transmitted to and stored in the blast amount data storage unit 11. The image data of the water surface of the reaction tank 2 taken by the camera 7 of FIG. 1 is transmitted to and stored in the image data storage unit 8. The image data transmitted and stored in the image data storage unit 8 is transmitted to the image processing unit 9. The image processing unit 9 processes image data by an operation described later, determines a bubble region on the water surface of the reaction tank, and calculates the area of the bubble region. The calculated bubble area is transmitted to and stored in the bubble area storage unit 10.

The air volume and bubble area data stored in the air volume data storage unit 11 and the bubble area storage unit 10 are transmitted to the regression model creation unit 12. The regression model creation unit 12 analyzes the data of these two systems, inputs the bubble area, and creates a regression model that outputs the air flow amount. The parameters of the created model are transmitted to and stored in the regression model parameter storage unit 12a.

When the regression model is created, the air volume estimation unit 13 reads out the regression model parameters from the regression model parameter storage unit 12a (arrow a), and based on the bubble area (arrow b) calculated from the newly acquired image data. Estimate the amount of air blown using the regression model created in. The estimated air volume is transmitted to the abnormality determination unit 14 (arrow c).

The abnormality determination unit 14 determines an abnormality based on the estimated value of the received air flow amount and the newly acquired actual air flow amount (arrow d), and the information on which the abnormality determination system 1 (not shown) is implemented. It is displayed on the display of the processing equipment (arrow e). To determine whether or not it is abnormal, the estimated air volume data (arrow c) and the actual air volume data (arrow d) are each provided with upper and lower thresholds, and whether or not the values are within the thresholds or estimated. The difference between the blown air volume data (arrow c) and the actual air volume data (arrow d) is calculated, the upper and lower threshold values are set for the value, and whether or not the difference is within the threshold value, or both of them are determined. Judgment may be made as a judgment criterion.

Next, the operation of the image processing unit 9 will be described in detail. The image processing unit 9 distinguishes the image data stored in the image data recording unit 8 into a water surface and bubbles floating on the water surface by image processing. The image processing unit 9 takes the sum of the number of pixels of the portion distinguished from the bubble, defines the sum as the area of the bubble, and makes it possible to quantify the amount of the bubble.

The reaction tank 2 is illuminated by lighting because it is photographed by the camera 7, but it is difficult to illuminate the water surface with uniform brightness, and there is a possibility that the lighting condition may differ. This problem is less affected by the lighting condition and makes it possible to distinguish between the water surface and bubbles by appropriately setting the threshold value using the adaptive binarization described below.

The procedure for quantifying bubbles will be described below.
(1) The acquired image of 3 channels of RGB is grayscaled and converted into 1 channel and a pixel value of 0 to 255.
(2) Apply a smoothing filter to remove noise.
(3) The water surface is converted to a pixel value of 0 (black) and bubbles are converted to a pixel value of 1 (white) by adaptive binarization.
(4) The area of white (bubbles) is calculated by summing the pixel values of the entire image.
Here, since the bubble has a pixel value of 1, the sum of the pixel values of the entire image expresses the area of the bubble.

Next, adaptive binarization will be described with reference to FIG. FIG. 3 is a diagram for explaining a procedure for processing an image taken by the camera 7. FIG. 3A is a grayscale image obtained through the above-mentioned bubble quantification procedures (1) and (2). As shown in FIG. 3B, in the adaptive binarization process, the pixel of interest (x, y) is a pixel located in the center of the pixel region of the filter (kxl). The threshold value σ of the pixel of interest (x, y) is set by the median value or the average value of the pixel values in the filter (kxl).

As shown in the following equation [Equation 1], when the pixel value I (x, y) of interest is larger than σ, the pixel value D (I (x, y)) of the binarized image is set to 1. When the pixel value I (x, y) of interest is σ or less, the pixel value D (I (x, y)) of the binarized image is set to 0.

FIG. 3C is a binary image converted by performing the above adaptive binarization process. In adaptive binarization, the filter size is set arbitrarily and the median or average value of the pixel values in the filter is set as the threshold value, so that it is not affected by the lighting condition and is relative in the filter. A bright area (foam) can be classified as white.

Next, the operation of the regression model creation unit 12 will be described in detail. In the regression model creation unit 12, the area (b _(tn) , ..., b _(t-1) ) of the nearest n bubbles in the past from the current time t and the air volume (a _(tn) , ..., Using the data of a _(t-1) ), a regression model f that estimates the amount of air blown by using the area of the bubble as an input is created. Here, the bubble area (b _(tn) , ..., B _(t-1) ) is data read from the bubble area storage unit 10 into the regression model creation unit 12. The blast volume (a _(tn) , ..., a _(t-1) ) is data read from the blast volume data storage unit 11 into the regression model creation unit 12.

When creating a regression model based on these data, it is also possible to take a moving average for the air volume and bubble area in order to reduce the effect of outliers and estimate the average value. FIG. 4 shows an image when data is converted by a moving average. The horizontal axis of FIG. 4 represents the passage of time, and the vertical axis represents the value of the air volume or the area of bubbles. The graph like a sawtooth wave shown by the line p is a straight line connecting the obtained data, and is greatly affected by noise. It can be seen that the broken line q is a moving average and is not easily affected by noise. By creating a regression model based on the value converted by this moving average, it is possible to create a highly accurate regression model with the influence of noise reduced. As a method for creating a regression model, a plurality of analytical means such as linear regression such as lasso regression and ridge regression, and nonlinear regression such as kernel regression can be mentioned.

In the blast amount estimation unit 13, the blast amount is estimated by the regression model f created by the regression model creation unit 12. Using the regression model equation _f with the amount bt of bubbles in the image for which the amount of air blown is to be estimated as an input, the estimated amount of air blown E (at) = _f ( _bt ) is obtained. Hereinafter, the above-mentioned abnormality determination is performed by the abnormality determination unit 14 based on this estimated value and the actual air flow amount.

As described above, in the present embodiment, the area of bubbles on the water surface is calculated from the image data obtained by photographing the water surface of the reaction tank 2, a regression model is created from the area of bubbles and the air flow amount data, and the regression model is used. The amount of air blown is estimated from the area of bubbles in the newly acquired image, and an abnormality is determined based on this estimated value and the actual amount of air blown. That is, since the abnormality determination of the processing device 100 can be performed based not only on the value of the air flow rate measuring device 6 but also on the multifaceted data, the accuracy of the abnormality determination can be improved. As a specific determination method, the difference between the estimated amount of air blown and the actual amount of air blown is taken, the difference is evaluated by a set threshold value, and an abnormality is determined. This makes it possible to improve the accuracy of the determination.

Further, by taking an image of the reaction tank 2 with the camera 7 and remotely monitoring it, the need to patrol the observer for on-site operation status management and abnormality confirmation can be reduced, and the burden on the worker can be reduced. .. Further, since the abnormality determination is performed based on the captured image data, it is possible to eliminate the oversight of the abnormality of the observer. By distinguishing the bubbles and the water surface of the reaction tank 2 by adaptive binarization and calculating the area of the bubbles, it is possible to quantitatively grasp the generation state of the bubbles.

(Second Embodiment)
FIG. 5 is a functional block diagram showing the configuration of the abnormality determination system 20 according to the present embodiment. The difference between the present embodiment and the first embodiment is that the abnormality determination system 20 includes a water quality data storage unit 15 for storing the water quality data of the reaction tank 2. Other components similar to those in the first embodiment are designated by the same reference numerals, and the description thereof will be omitted.

The amount (area) of bubbles in the reaction tank is not only determined by the amount of air blown, but is also affected by the water quality. In the first embodiment, the amount of air blown is estimated only by the area of bubbles, whereas in the present embodiment, the water quality data from the sensor of the reaction tank 2 is stored in the newly added water quality data storage unit 15, and these information are also stored. use. The water quality data includes activated sludge suspended matter, excess sludge concentration, inflow flow rate, ammonium nitrogen, chemical oxygen demand, biological oxygen demand, total nitrogen, total phosphorus, and nitrite nitrogen measured in the reaction tank 2. , Nitrate nitrogen, phosphate phosphorus, etc.

In the present embodiment, the regression model creating unit 12 creates a regression model that estimates the amount of air blown by inputting the amount of bubbles (area of bubbles) and water quality data. The regression model parameter storage unit 12a stores the model parameters obtained by the regression model creation unit 12. The blast amount estimation unit 13 creates a regression model from the parameters stored in the regression model parameter storage unit 12a, and estimates the blast amount from the newly acquired bubble amount and water quality data. For the following processing, the same processing as in the first embodiment is performed, and the state of the processing apparatus 100 is determined.

In this embodiment, in addition to the same action and effect as in the first embodiment, a regression model is created using the amount of bubbles in the reaction tank 2 and the water quality data, so that the amount of air blown can be estimated more accurately.

(Third Embodiment)
FIG. 6 is a functional block diagram showing the configuration of the abnormality determination system 30 according to the present embodiment. This embodiment differs from the second embodiment in that it includes a deep learning model creating unit 16 and a deep learning model parameter storage unit 16a that use deep learning as a means for estimating the amount of air blown.
Other components similar to those in the second embodiment are designated by the same reference numerals, and the description thereof will be omitted.

In the first embodiment, the area of bubbles calculated by image processing is used, and in the second embodiment, the amount of air blown is estimated using water quality data in addition to the area of bubbles. Therefore, two steps are required for image processing and regression model creation. The image processing unit 9 needs a filter size for adaptive binarization as a parameter. When creating a regression model in the regression model creating unit 12, there is a possibility that trial and error of parameter adjustment may be required to improve the accuracy, which may be a burden on the user.

Since the deep learning method is used in this embodiment, the image processing unit 9 and the regression model creation unit 12 of the first embodiment and the second embodiment are replaced, and the deep learning model creation unit 16 corresponds to the input information and the amount of air blown. Automatically tune parameters from. Although there are parameters for filter size and network structure in deep learning, the burden of trial and error on the user is reduced by using the recommended values that have been reported to exhibit good accuracy in many cases. And estimate the air volume accurately.

In the following, the operation outline of FIG. 6 will be described with respect to the portion newly added from the second embodiment.
The deep learning model creation unit 16 inputs the pixel value of the image data and the water quality data, and creates a deep learning model for estimating the amount of air blown. At this time, it is also possible to apply a filter to the image to perform the convolution process. An example of a deep learning model is shown in FIG. The feature amount of the image data of 3 channels is extracted by the convolution layer and the max pooling layer S. The feature amount of the extracted image data and the water quality data such as ammonium nitrogen and phosphoric acid phosphorus are put into the fully bonded layer U, and the weight parameter of deep learning for estimating the air flow amount is adjusted. The deep learning model parameter storage unit 16a stores the parameters created by the deep learning model creation unit 16. The blast amount estimation unit 13 creates a deep learning model from the parameters stored in the deep learning model parameter storage unit 16a, and estimates the blast amount from the newly acquired bubble amount and water quality data. For the following processing, the same processing as in the first embodiment is performed, and the state of the processing apparatus 100 is determined.

In the present embodiment, in addition to the operation and effect of the second embodiment, the image processing unit 9 and the regression model creation unit 12 of the second embodiment are replaced with the deep learning model creation unit 16, so that the parameters and regression of the image processing can be obtained. The burden of proper setting of the model can be removed. By automatically tuning the parameters of the deep learning model from the correspondence between the input information (image and water quality data) and the output information (air volume), the air volume can be estimated accurately.

1 Abnormality determination system 2 Reaction tank 3 Blower 8 Image data storage unit 9 Image processing unit 10 Bubble area storage unit 11 Blower volume data storage unit 12 Regression model creation unit 13 Blower volume estimation unit 14 Abnormality determination unit 15 Water quality data storage unit 16 Deep layer Learning model creation unit 100 Processing device

Claims (5)

It is an abnormality judgment system of a processing device equipped with a reaction tank and a blower.
An image data storage unit that stores image data obtained by photographing the reaction tank, and
An image processing unit that calculates the area of bubbles in the previous image data,
A bubble area storage unit that stores the area of the bubbles,
An air volume data storage unit that stores the air volume of the blower, and
A regression model creation unit that creates a regression model based on the area of the bubbles and the amount of air blown.
An air volume estimation unit that estimates the air volume from the newly acquired bubble area and the regression model,
An abnormality determination system including an abnormality determination unit that determines an abnormality based on the estimated air flow amount and the actual air flow amount. A water quality data storage unit for storing the water quality data of the reaction tank is provided.
The regression model creation unit creates a regression model based on the water quality data, the area of the bubbles, and the amount of air blown.
The abnormality determination system according to claim 1, wherein the blast amount estimation unit estimates the blast amount from the newly acquired bubble area and water quality data and the regression model. It is an abnormality judgment system of a processing device equipped with a reaction tank and a blower.
An image data storage unit that stores image data obtained by photographing the reaction tank, and
An air volume data storage unit that stores the air volume of the blower, and
A water quality data storage unit that stores the water quality data of the reaction tank,
A deep learning model creation unit that creates a deep learning model based on the water quality data, the image data, and the air volume.
An airflow amount estimation unit that estimates the airflow amount from the newly acquired image data and water quality data, and the deep learning model.
An abnormality determination system including an abnormality determination unit that determines an abnormality based on the estimated air flow amount and the actual air flow amount. The abnormality determination system according to claim 1 or 2, wherein the image processing unit calculates the area of bubbles by adaptive binarization. It is a method of determining an abnormality in a processing device equipped with a reaction tank and a blower.
Taking a picture of the reaction tank,
Calculating the area of bubbles in the image data taken in the previous term,
Measuring the amount of air blown by the blower,
Creating a regression model based on the area of the bubbles and the amount of air blown,
Estimating the amount of air blown from the area of bubbles in the newly captured image data and the regression model,
An abnormality determination method including determining an abnormality based on the estimated air flow amount and the actual air flow amount.

Images