Maintenance timing determination device, maintenance timing determination method, and maintenance timing determination program

Abstract

To provide a maintenance timing determination apparatus capable of determining the timing when an instrumentation sensor used for a water treatment device is to be washed or cleaned, more accurately.SOLUTION: A maintenance timing determination apparatus includes a data acquisition unit, an inference unit, and a maintenance processing unit. The data acquisition unit acquires measurement value data measured by an instrumentation sensor, a post-washing time elapsed since washing was completed, post-cleaning time elapsed since cleaning was completed, and the number of times of cleaning. The inference unit outputs maintenance timing from the measurement value data, the post-washing time, the post-cleaning time, and the number of times of cleaning input from the data acquisition unit, using a maintenance timing inference model for inferring the maintenance timing for one of no-maintenance, washing, and cleaning, from the measurement data, the post-washing time, the post-cleaning time, and the number of times of cleaning. The maintenance processing unit performs processing in accordance with the output maintenance timing.SELECTED DRAWING: Figure 3

Description

[Technical Field]

[0001] The present disclosure relates to a maintenance timing determination device, a maintenance timing determination method, and a maintenance timing determination program that determine the timing of washing or cleaning an instrumentation sensor provided in a water treatment device. [Background technology]

[0002] Conventionally, techniques for determining the timing of maintenance of sensors that measure the status of a plant have been known. Patent Document 1 discloses a technique for generating a judgment model that uses data that associates measurement data acquired from the sensor with maintenance information indicating the content of maintenance work performed on the sensor after acquiring the measurement data to determine whether there is a problem or whether maintenance work is required from the sensor measurement data. Examples of sensor maintenance work include zero point adjustment, sensor cleaning, and sensor replacement. Patent Document 1 also discloses a technique that inputs data acquired from the sensor into the generated judgment model to determine whether there is a problem with the sensor or which maintenance work is required, among zero point adjustment, sensor cleaning, and sensor replacement, and, if maintenance work is required, generates a sensor maintenance plan based on the determination result. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Publication No. 2019-220226 Summary of the Invention [Problem to be solved by the invention]

[0004] Incidentally, instrumentation sensors used to measure the concentration of pollutants in water treatment systems such as sewage treatment plants often undergo two procedures: cleaning, in which cleaning water is sprayed onto the instrumentation sensor from a cleaning machine installed near the sensor's installation position in the piping, and manual cleaning, in which the sensor is removed from the piping. However, sensor cleaning, which is one of the sensor maintenance tasks in the above-mentioned conventional technology, corresponds to the latter cleaning, and does not distinguish between cleaning and washing. Furthermore, even in the above-mentioned conventional technology, it is possible to include cleaning and washing in maintenance tasks. However, there is a problem in that it is difficult to accurately determine whether cleaning or washing should be performed simply by using physical quantity data such as pressure, temperature, pH, and product flow rate as measurement data, as exemplified in the above-mentioned conventional technology.

[0005] The present disclosure has been made in consideration of the above, and aims to provide a maintenance timing determination device that can determine the timing to wash or clean instrumentation sensors used in water treatment devices with greater accuracy than conventional methods. [Means for solving the problem]

[0006] To solve the above-mentioned problems and achieve the object, the maintenance timing determination device according to the present disclosure is provided in a water treatment device and determines the timing of maintenance for an instrumentation sensor that measures pollutants in water to be treated. The maintenance timing determination device includes a data acquisition unit, an inference unit, and a maintenance processing unit. The data acquisition unit acquires measurement value data measured by the instrumentation sensor at one or more times, a post-cleaning elapsed time that is the elapsed time since the instrumentation sensor was cleaned by a cleaning machine, a post-cleaning elapsed time that is the elapsed time since cleaning performed by removing the instrumentation sensor was completed, and a cleaning count that is the number of times cleaning was determined. The inference unit uses a maintenance timing inference model, which is a trained model for inferring the timing of maintenance (whether no maintenance is required or whether cleaning is required) from the measurement value data, the post-cleaning elapsed time, the post-cleaning elapsed time, and the cleaning count, to output the maintenance timing from the measurement value data, the post-cleaning elapsed time, and the cleaning count input from the data acquisition unit. The maintenance processing unit performs processing according to the maintenance timing output from the inference unit. [Effects of the Invention]

[0007] According to the present disclosure, it is possible to more accurately determine the timing to wash or clean an instrumentation sensor used in a water treatment device than in the past. [Brief explanation of the drawings]

[0008] [Figure 1] FIG. 1 is a diagram showing an example of the configuration of a water treatment system including a maintenance timing determination device according to a first embodiment. [Figure 2] A diagram showing an example of the positional relationship between an instrumentation sensor and a cleaning machine in a piping system. [Figure 3] FIG. 1 is a diagram schematically illustrating an example of the configuration of a maintenance timing determination device according to a first embodiment. [Figure 4] FIG. 1 is a diagram schematically illustrating an example of a neural network used by a model generation unit. [Figure 5] Flowchart showing an example of a procedure for a maintenance method [Figure 6] Flowchart showing an example of the procedure for generating a trained model [Figure 7] A flowchart showing an example of a procedure for determining the timing of maintenance [Figure 8] FIG. 10 is a diagram schematically illustrating another example of the configuration of the maintenance timing determination device according to the first embodiment. [Figure 9] FIG. 10 is a diagram illustrating an example of the configuration of an instrumentation device connected to a maintenance timing determination device according to a second embodiment. [Figure 10] FIG. 1 is a block diagram showing an example of the configuration of a computer system that realizes a maintenance timing determination device according to first and second embodiments. DETAILED DESCRIPTION OF THE INVENTION

[0009] DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS A maintenance timing determination device, a maintenance timing determination method, and a maintenance timing determination program according to embodiments of the present disclosure will be described in detail below with reference to the accompanying drawings.

[0010] Embodiment 1 The maintenance timing determination device according to the first embodiment is a device that determines the maintenance timing, which is the timing for automatic cleaning by a cleaning machine or cleaning by an operator, of an instrumentation sensor that detects the concentration of pollutants contained in water to be treated in a water treatment device. The operator includes an operations manager of the water treatment system. Examples of water treatment devices are water purification plants, sewage treatment plants, and factory wastewater treatment facilities. Below, the determination of the maintenance timing of an instrumentation sensor that detects the concentration of activated sludge, which is an example of pollutants, will be described using a case where the water treatment device is a sewage treatment plant as an example.

[0011] FIG. 1 is a diagram showing an example of the configuration of a water treatment system including a maintenance timing determination device according to a first embodiment. The water treatment system 100 is a system that converts wastewater such as sewage into purified treated water using biological purification technology using activated sludge. Hereinafter, the wastewater will be referred to as water to be treated. The water treatment system 100 includes a water treatment device 101 that mixes the water to be treated with activated sludge to obtain purified treated water.

[0012] The water treatment device 101 includes an anoxic tank 2, an aerobic tank 3, and a final settling tank 4. The anoxic tank 2 is a tank that receives the water to be treated. The aerobic tank 3 is a tank that receives the anoxic tank-treated liquid, which is the treated liquid that flows out from the anoxic tank 2. The final settling tank 4 is a tank that receives the aerobic tank-treated liquid, which is the treated liquid that flows out from the aerobic tank 3, and separates the activated sludge contained in the aerobic tank-treated liquid by solid-liquid separation to obtain treated water. The water treatment device 101 also includes a nitrification solution circulating pump 5, a sludge withdrawal pump 9, instrumentation sensors 11a and 11b, and washers 12a and 12b. The nitrification solution circulating pump 5 sends the activated sludge remaining in the aerobic tank 3 to the anoxic tank 2. The sludge withdrawal pump 9 withdraws the activated sludge that has accumulated at the bottom of the final settling tank 4.

[0013] An influent pipe 1 is connected to the anoxic tank 2, and the water to be treated flows into the anoxic tank 2 through the influent pipe 1. The anoxic tank 2 has a submersible agitator 8. The submersible agitator 8 mixes the activated sludge remaining in the anoxic tank 2 with the water to be treated. In other words, the water to be treated is treated by the activated sludge in an anoxic environment, i.e., in a state where the molecular oxygen concentration is extremely low. Specifically, the nitrate ions (NO3 - ) is reduced by microbial action and converted into nitrogen gas, which is then removed from the water. This denitrification process takes place in the anoxic tank 2. The nitrified liquid is the activated sludge mixed liquid that has been retained in the aerobic tank 3. The activated sludge mixed liquid that flows out of the anoxic tank 2 flows into the aerobic tank 3.

[0014] The aerobic tank 3 is provided at the bottom with an aeration device 6 that supplies oxygen to the activated sludge mixed liquid retained in the aerobic tank 3, and a blower 7 that pressure-feeds oxygen-containing gas such as air to the aeration device 6 through piping. In the aerobic tank 3, the water to be treated is treated under aerobic conditions. Specifically, the ammonium ions (NH4 + Nitrification, in which nitrification ions are converted into nitrate ions by the action of microorganisms, takes place in the aerobic tank 3. As described above, the aerobic tank 3 is connected to the nitrification liquid circulation pump 5. The nitrification liquid circulation pump 5 extracts a portion of the nitrification liquid, which is the activated sludge mixed liquid retained in the aerobic tank 3, and returns it to the anoxic tank 2. The activated sludge mixed liquid flowing out of the aerobic tank 3 flows into the final settling tank 4.

[0015] The final settling tank 4 separates the activated sludge mixed liquor from the aerobic tank 3 into solid and liquid. Specifically, the activated sludge from the inflowing activated sludge mixed liquor settles and separates to the bottom of the final settling tank 4 by gravity, and the supernatant water flows out from the top of the final settling tank 4 and is sent as treated water to subsequent treatment stages such as chlorination. A sludge withdrawal pump 9 is connected to the final settling tank 4. The sludge withdrawal pump 9 extracts some of the activated sludge that has accumulated at the bottom of the settling tank and returns it to the anoxic tank 2.

[0016] The instrumentation sensor 11a is provided in a pipe, which is an example of a flow path that sends activated sludge mixed liquid from the aerobic tank 3 to the anoxic tank 2, and measures the concentration of activated sludge in the water to be treated flowing through the pipe. The instrumentation sensor 11a outputs the measurement results to a maintenance timing determination device 20, which will be described later. The instrumentation sensor 11b is provided in a pipe that sends activated sludge from the final settling tank 4 to the anoxic tank 2, and measures the concentration of activated sludge in the water to be treated flowing through the pipe. The instrumentation sensor 11b outputs the measurement results to a maintenance timing determination device 20, which will be described later. One example of the instrumentation sensors 11a and 11b is a scattered light sludge concentration meter that irradiates the water to be treated with light such as a laser beam and measures the intensity of the scattered light scattered by the activated sludge in the water to be treated to measure the concentration of activated sludge. Hereinafter, the instrumentation sensors 11a and 11b will be referred to as instrumentation sensor 11 when not being distinguished from one another.

[0017] The cleaning machine 12a sprays cleaning water onto the surface of the instrumentation sensor 11a to clean the surface of the instrumentation sensor 11a. The cleaning machine 12b sprays cleaning water onto the surface of the instrumentation sensor 11b to clean the surface of the instrumentation sensor 11b. The cleaning machines 12a and 12b clean the surfaces of the instrumentation sensors 11a and 11b in accordance with instructions from a maintenance timing determination device 20, which will be described later. Hereinafter, the cleaning machines 12a and 12b will be referred to as the cleaning machine 12 when not being individually distinguished.

[0018] 2 is a diagram showing an example of the positional relationship between the instrumentation sensor and the washer in the piping. The instrumentation sensor 11 is inserted into and fixed in a through-hole 151 provided in the side of the piping 15. In one example, the instrumentation sensor 11 is fixed to the piping 15 with its tip protruding beyond the inner wall of the piping 15. If the part of the surface of the tip of the instrumentation sensor 11 that detects activated sludge is defined as a detection area 111, the instrumentation sensor 11 is attached so that the detection area 111 faces the inside of the piping 15.

[0019] The cleaning machine 12 is provided at a position where it can clean the detection area 111 of the instrumentation sensor 11. The cleaning machine 12 has an injection port 121 at its tip for injecting cleaning water, and the injection port 121 is provided so as to face the detection area 111 of the instrumentation sensor 11. The injection of cleaning water by the cleaning machine 12 onto the detection area 111 of the instrumentation sensor 11 is called cleaning. The cleaning machine 12 has a valve 122 that switches between injecting and not injecting cleaning water. The opening and closing of the valve 122 is controlled by a control signal from the maintenance timing determination device 20.

[0020] Activated sludge adheres to the detection area 111 of the instrumentation sensor 11. When activated sludge adheres to the detection area 111, it is cleaned by the washer 12. However, cleaning by the washer 12 does not completely remove the activated sludge. Therefore, even if cleaning is performed, more activated sludge gradually accumulates on the attached activated sludge, and the activated sludge becomes stuck to the detection area 111. Since the stuck activated sludge cannot be removed by cleaning by the washer 12, it is removed by cleaning by an operator. In FIG. 2, the instrumentation sensor 11 is detachable from the piping 15. Therefore, when cleaning is performed, the instrumentation sensor 11 is removed under conditions where the water to be treated does not flow through the piping 15, and the stuck activated sludge is removed by an operator. Note that, hereinafter, activated sludge attached to the instrumentation sensor 11 is also referred to as dirt, and the state in which activated sludge adheres or sticks to the instrumentation sensor 11 is also referred to as "fouling."

[0021] If contamination adheres to the detection area 111 of the instrumentation sensor 11, light from the instrumentation sensor 11 will not be properly irradiated onto the water being treated in the pipe 15, and it will be difficult to properly measure the scattered light intensity in the water being treated. If the treatment in the water treatment device 101 is controlled based on the activated sludge concentration measured by the instrumentation sensor 11, a discrepancy greater than the margin of error will occur between the actual activated sludge concentration in the water being treated and the concentration measured by the instrumentation sensor 11. As a result, the treatment in the water treatment device 101 will not be as intended. To avoid this situation, washing or cleaning using the washer 12 is required so that the instrumentation sensor 11 can properly measure the activated sludge concentration in the water being treated.

[0022] Returning to Figure 1, water treatment system 100 includes a maintenance timing determination device 20 that determines the timing of maintenance for instrumentation sensor 11 using information including the results of measurements by instrumentation sensor 11. Using the measurement results from instrumentation sensor 11, maintenance timing determination device 20 determines whether a change in the measurement value by instrumentation sensor 11 is a normal change in the measurement value due to a change in the concentration of activated sludge in the water to be treated, or an abnormal change in the measurement value due to contamination of detection area 111 preventing normal measurement, and if the change in the measurement value is abnormal, determines whether washing or cleaning should be performed. In other words, maintenance timing determination device 20 infers whether the maintenance timing is no longer required, or whether it is washing or cleaning.

[0023] 3 is a diagram schematically illustrating an example of the configuration of the maintenance timing determination device according to Embodiment 1. The maintenance timing determination device 20 includes a cleaning determination timer 21, a re-cleaning determination timer 22, a cleaning counting unit 23, a data acquiring unit 24, a trained model storage unit 25, an inference unit 26, a maintenance processing unit 27, and a display unit 28.

[0024] The cleaning determination timer 21 measures the post-cleaning elapsed time, which is the time elapsed since the cleaning of the instrumentation sensor 11 by the cleaning machine 12 is completed. When an abnormal change in the measurement value is detected after the completion of cleaning, the cleaning determination timer 21 measures the post-cleaning elapsed time to determine whether to perform cleaning or cleaning. The cleaning determination period T1 is the period that serves as the basis for determining whether to perform cleaning or cleaning, i.e., the period during which it is determined that cleaning will be performed rather than cleaning. In other words, if an abnormal change in the measurement value is detected during the cleaning determination period T1 after the completion of cleaning, cleaning will be performed. In one example, the cleaning determination period T1 is set by an operator. The cleaning determination timer 21 starts measuring the post-cleaning elapsed time after receiving a cleaning completion signal from the maintenance processing unit 27, which is a signal indicating the completion of cleaning by the cleaning machine 12. Note that the abnormal change in the measurement value corresponds to an abnormality in the state of the instrumentation sensor 11.

[0025] The re-cleaning determination timer 22 measures the post-cleaning elapsed time, which is the time elapsed since the completion of cleaning with the instrumentation sensor 11 removed. When an abnormal change in measurement value is detected after the completion of cleaning, the re-cleaning determination timer 22 measures the post-cleaning elapsed time to determine whether to perform re-cleaning or cleaning. The re-cleaning determination period T2 is the criterion for determining whether to perform re-cleaning or cleaning. In other words, if an abnormal change in measurement value is detected during the re-cleaning determination period T2 after the completion of cleaning, it is determined that cleaning of the instrumentation sensor 11 should be performed again. Note that if an abnormal change in measurement value is detected again during the re-cleaning determination period T2 after the completion of cleaning, it is highly likely that an abnormality has occurred in the instrumentation sensor 11. In one example, a period during which abnormal changes in measurement value are not normally expected to occur after the completion of cleaning is calculated from past data and set as the re-cleaning determination period T2. In one example, the re-cleaning determination period T2 is set by an operator. In one example, the re-cleaning determination timer 22 starts counting the time that has elapsed since cleaning after receiving a cleaning completion signal, which is a signal indicating the completion of cleaning, from an information processing terminal held by the worker.

[0026] The cleaning count counter 23 counts the number of cleanings, which is the number of times a cleaning is determined to have been performed within a predetermined period of time after the cleaning was performed. Specifically, the cleaning count counter 23 counts the first cleaning when an abnormal change in the measurement value is detected while the re-cleaning determination timer 22 is not timing, and thereafter increments the count each time the re-cleaning determination timer 22 detects an abnormal change in the measurement value while timing. In addition, the cleaning count counter 23 resets the count when a predetermined re-cleaning determination period T2 has elapsed since the re-cleaning determination timer 22 started timing.

[0027] When inferring the timing of maintenance, the data acquisition unit 24 acquires data necessary to determine the timing of maintenance. The data necessary to determine the timing of maintenance includes the measurement value data of the instrumentation sensor 11, the time elapsed since cleaning which is a value measured by the cleaning determination timer 21, the time elapsed since cleaning which is a value measured by the re-cleaning determination timer 22, and the number of cleanings counted by the cleaning count counter 23.

[0028] The measurement value data includes measurement values ​​that are measurement results from the instrumentation sensor 11. The measurement value data includes measurement values ​​measured by the instrumentation sensor 11 at one or more times. The measurement value data is used to determine whether a change in the measurement value from the instrumentation sensor 11 is due to a change in the concentration of activated sludge in the water to be treated, an abnormality in the measurement value due to dirt adhering to the instrumentation sensor 11, or an abnormality in the instrumentation sensor 11 itself. The measurement value data preferably includes multiple measurement values ​​over a set period of time, but may also include one measurement value at a certain time.

[0029] The post-cleaning elapsed time is data indicating a value measured by the cleaning determination timer 21. The post-cleaning elapsed time is output from the cleaning determination timer 21. The post-cleaning elapsed time is data indicating a value measured by the re-cleaning determination timer 22. The post-cleaning elapsed time is output from the re-cleaning determination timer 22. The number of cleanings is data indicating the number of times an abnormal change in measurement value occurs during the re-cleaning determination period T2. The number of cleanings is output from the cleaning count unit 23.

[0030] The trained model storage unit 25 stores a maintenance timing inference model, which is a trained model for inferring maintenance timing from the measurement value data, the time elapsed since cleaning, the time elapsed since cleaning, and the number of cleanings. The maintenance timing includes at least whether maintenance is unnecessary, whether the instrumentation sensor 11 needs to be cleaned by the cleaning machine 12, or whether the instrumentation sensor 11 needs to be cleaned.

[0031] The inference unit 26 infers the maintenance timing obtained by using the maintenance timing inference model, which is a trained model. That is, by inputting the measurement value data, the time elapsed since cleaning, the time elapsed since cleaning, and the number of cleanings acquired by the data acquisition unit 24 into this maintenance timing inference model, it is possible to output the maintenance timing inferred from the measurement value data, the time elapsed since cleaning, the time elapsed since cleaning, and the number of cleanings. The inference unit 26 outputs the inference result to the maintenance processing unit 27. In this example, the inference result is either no maintenance required, or cleaning and cleaning.

[0032] When the maintenance processing unit 27 receives the maintenance timing as an inference result from the inference unit 26, it performs processing according to the maintenance timing. If the inference result indicates that maintenance is not required, the maintenance processing unit 27 does not perform any special processing to maintain the current state.

[0033] If the inference result is cleaning, the maintenance processing unit 27 causes the display unit 28 to display cleaning alarm information indicating that cleaning by the cleaning machine 12 is necessary. The maintenance processing unit 27 also outputs a control signal to instruct cleaning to the cleaning machine 12. After cleaning by the cleaning machine 12 is completed, the maintenance processing unit 27 causes the cleaning determination timer 21 to start counting and resets the count in the cleaning count count unit 23.

[0034] If the inference result is cleaning, the maintenance processing unit 27 displays cleaning alarm information indicating that cleaning is required on the display unit 28. The maintenance processing unit 27 increments the cleaning count in the cleaning count counter 23 by "1." The worker sees the cleaning alarm information displayed on the display unit 28, stops the operation of the water treatment device 101, and then performs cleaning. When cleaning is complete, the worker transmits a cleaning completion signal from the information processing terminal to the maintenance timing determination device 20. When the maintenance processing unit 27 receives the cleaning completion signal from the information processing terminal held by the worker after cleaning is completed, it causes the re-cleaning determination timer 22 to start counting. Note that if the cleaning count in the cleaning count counter 23 is "2" or more, the maintenance processing unit 27 may include in the cleaning alarm information the number of consecutive notifications of the cleaning alarm information indicating that cleaning is required within the re-cleaning determination period T2. The display unit 28 displays the number of consecutive cleaning alarm notifications, allowing the worker to recognize the possibility of an abnormality in the instrumentation sensor 11. That is, when cleaning again, it is possible to not only simply clean the instrumentation sensor 11 but also to check whether there is any abnormality in the instrumentation sensor 11.

[0035] The display unit 28 displays the washing alarm information or cleaning alarm information in accordance with instructions from the maintenance processing unit 27. It is desirable that the washing alarm information or cleaning alarm information be displayed in a manner that can attract the attention of the operator.

[0036] The maintenance timing inference model used by the inference unit 26 can be updated or generated by the maintenance timing determination device 20. Fig. 3 shows a case where a maintenance timing inference model can be generated by the maintenance timing determination device 20. That is, the maintenance timing determination device 20 includes a learning data storage unit 29 and a model generation unit 30.

[0037] When generating a maintenance timing inference model, the data acquisition unit 24 acquires data necessary for training the maintenance timing inference model and stores it in the training data storage unit 29. The data necessary for training the maintenance timing inference model is training data that combines measurement value data, the elapsed time since cleaning, the elapsed time since cleaning, the number of cleanings, and the maintenance content judgment result at that time. The maintenance content judgment result is the result of determining whether maintenance is not required or whether cleaning and cleaning are required based on the state of the instrumentation sensor 11 when the measurement value data, the elapsed time since cleaning, the elapsed time since cleaning, and the number of cleanings are acquired. The judge is preferably an experienced worker. The judge determines the maintenance content judgment result by taking into account whether the elapsed time since cleaning is within a specified cleaning judgment period T1, whether the elapsed time since cleaning is within a specified re-cleaning judgment period T2, and the number of cleanings. The maintenance content judgment result may be a combination of the result inferred by the inference unit 26 and the judgment result of an experienced worker. Here, an experienced worker is, for example, a worker who is familiar with work in the water treatment system 100 and who, based on experience, is able to observe changes in the measurement values ​​of the instrumentation sensor 11 and the state of the detection area 111 of the instrumentation sensor 11 and infer from experience whether maintenance is not required or whether it needs to be washed or cleaned.

[0038] The learning data storage unit 29 stores a plurality of learning data sets that correspond to data sets including measurement value data, time elapsed since cleaning, time elapsed since cleaning, and number of cleanings, and correct answer data including the maintenance content judgment result at that time.

[0039] The model generation unit 30 learns the maintenance timing based on the learning data created based on a combination of a data set including the measurement value data, the elapsed time since cleaning, the elapsed time since cleaning, and the number of cleanings acquired by the data acquisition unit 24, i.e., stored in the learning data storage unit 29, and the maintenance content judgment result, which is the correct answer data at that time. That is, a maintenance timing inference model is generated, which is a trained model that infers the optimal maintenance timing from the data set including the measurement value data of the instrumentation sensor 11, the elapsed time since cleaning, the elapsed time since cleaning, and the number of cleanings, and the maintenance content judgment result at that time. Here, the learning data is data that correlates the data set including the measurement value data, the elapsed time since cleaning, the elapsed time since cleaning, and the number of cleanings with the maintenance content judgment result at that time. Furthermore, the maintenance timing inference model uses the elapsed time since cleaning and the elapsed time since cleaning when inferring the maintenance timing.

[0040] The learning algorithm used by the model generation unit 30 can be a known algorithm such as supervised learning. As an example, a case where a neural network is applied will be described. Note that it is also possible to generate a trained model, i.e., a maintenance timing inference model, using a learning method other than that shown here.

[0041] In one example, the model generation unit 30 learns the maintenance timing according to a neural network model using so-called supervised learning. Here, supervised learning refers to a technique in which a learning device is provided with pairs of input and result label data, and the learning device learns the features of the learning data and infers the result from the input.

[0042] A neural network consists of an input layer consisting of multiple neurons, an intermediate layer consisting of multiple neurons, and an output layer consisting of multiple neurons. The intermediate layer, also called a hidden layer, can be one layer or two or more layers.

[0043] FIG. 4 is a diagram schematically illustrating an example of a neural network used by the model generation unit. For example, in a three-layer neural network such as that shown in FIG. 4, when multiple inputs are input to input layer X1 through input layer X3, the values ​​are multiplied by weights indicated by w11 through w16 and then input to intermediate layer Y1 through intermediate layer Y2. When weights w11 through w16 are not individually distinguished, they are referred to as weight w1. Furthermore, the results from intermediate layer Y1 through intermediate layer Y2 are further multiplied by weights indicated by w21 through w26 and output from output layer Z1 through output layer Z3. When weights w21 through w26 are not individually distinguished, they are referred to as weight w2. The output results from output layer Z1 through output layer Z3 vary depending on the values ​​of weights w1 and w2.

[0044] In the first embodiment, the neural network learns the maintenance timing by so-called supervised learning in accordance with the learning data created based on a combination of the data set and the correct answer data acquired by the data acquiring unit 24.

[0045] That is, the neural network learns by inputting a data set to the input layer and adjusting the weights w1 and w2 so that the result output from the output layer approaches the correct data.

[0046] The model generation unit 30 generates and outputs a trained model by performing the above-described learning. The model generation unit 30 stores the generated trained model in the trained model storage unit 25.

[0047] 1, the water treatment device 101 includes a plurality of instrumentation sensors 11a and 11b. Therefore, the data acquisition unit 24 classifies the acquired data for each of the instrumentation sensors 11a and 11b. The model generation unit 30 may generate a maintenance timing inference model for each of the instrumentation sensors 11a and 11b using the learning data classified for each of the instrumentation sensors 11a and 11b. In this case, the inference unit 26 acquires the maintenance timing by inputting the data classified for each of the instrumentation sensors 11a and 11b into the maintenance timing inference model generated for each of the instrumentation sensors 11a and 11b.

[0048] Here, the data used to determine the timing of maintenance will be described. Fig. 5 is a flowchart showing an example of the procedure of the maintenance method. Fig. 5 shows an example of the processing procedure for determining the timing of maintenance using the above data.

[0049] First, measurement value data is acquired (step S11), and it is determined from the measurement value data whether an abnormal change in the measurement value has occurred (step S12). The presence or absence of an abnormal change in the measurement value can be determined using the measurement value data from the instrumentation sensor 11. Also, by using time-series data for a certain period to acquire the gradient of the activated sludge concentration over time, the presence of high-frequency components, and the like, it is possible to determine whether or not an abnormal change in the measurement value has occurred. In this way, the measurement value data from the instrumentation sensor 11 is acquired in order to detect the occurrence of an abnormal change in the measurement value.

[0050] If no abnormal change in the measurement value has occurred (No in step S12), the process returns to step S11. If an abnormal change in the measurement value has occurred (Yes in step S12), it is determined whether the elapsed time since cleaning, which is the value of the re-cleaning determination timer 22, is within the re-cleaning determination period T2 (step S13). The re-cleaning determination timer 22 starts timing after the previous cleaning is completed. If an abnormal change in the measurement value occurs immediately after cleaning, it is considered that there may be an abnormality in the instrumentation sensor 11. In other words, a state in which cleaning is determined to have occurred multiple times during the re-cleaning determination period T2 is unlikely to be due to adhesion of activated sludge, etc., and is considered to be an abnormality in the instrumentation sensor 11 itself, such as a malfunction. In this way, obtaining the elapsed time since cleaning is effective in detecting an abnormality in the instrumentation sensor 11 itself.

[0051] If the time elapsed since cleaning is not within the re-cleaning determination period T2 (No in step S13), it is determined whether the time elapsed since cleaning, which is the value of the cleaning determination timer 21, is within the cleaning determination period T1 (step S14). The cleaning determination timer 21 starts timing after the previous cleaning by the cleaning machine 12 is completed. If an abnormal change in the measurement value occurs immediately after cleaning, it is considered that activated sludge has already adhered to the detection area 111 of the instrumentation sensor 11 and cannot be removed by cleaning by the cleaning machine 12. In this way, obtaining the time elapsed since cleaning is effective in detecting the timing of cleaning rather than cleaning.

[0052] If the time elapsed since cleaning is not within the cleaning determination period T1 (No in step S14), the number of cleanings counted by cleaning count unit 23 is reset (step S15), and cleaning alarm information is output (step S16). In addition, cleaning machine 12 cleans instrumentation sensor 11 (step S17). When cleaning is completed, cleaning determination timer 21 starts counting the time elapsed since cleaning (step S18). Then, the process returns to step S11.

[0053] If the time elapsed since cleaning is within the cleaning determination period T1 in step S14 (Yes in step S14), or if the time elapsed since cleaning is within the re-cleaning determination period T2 in step S13 (Yes in step S13), the cleaning count of the cleaning count unit 23 is incremented by "1" (step S19). Then, cleaning alarm information is output (step S20). At this time, if the number of cleanings is "2" or more, information indicating that the cleaning alarm information has been output multiple times consecutively during the re-cleaning determination period T2 may be included in the cleaning alarm information. Then, after the operation of the water treatment device 101 is stopped by the operator, the instrumentation sensor 11 is cleaned (step S21). When the cleaning is completed, the re-cleaning determination timer 22 starts counting the time elapsed since cleaning (step S22). Then, the process returns to step S11.

[0054] As described above, to determine whether the instrumentation sensor 11 does not require maintenance or needs to be washed and cleaned, at least the measurement data of the instrumentation sensor 11, the time elapsed since cleaning, the time elapsed since cleaning, and the number of cleanings are required. Therefore, in the first embodiment, when inferring the timing for washing and cleaning maintenance of the instrumentation sensor 11, the measurement data of the instrumentation sensor 11, the time elapsed since cleaning, the time elapsed since cleaning, and the number of cleanings are used.

[0055] Next, a method for generating a trained model and a method for determining maintenance timing in the maintenance timing determination device 20 will be described.

[0056] <How to generate a trained model> 6 is a flowchart showing an example of the steps of a method for generating a trained model. The data acquisition unit 24 acquires a data set including measurement value data, the time elapsed since cleaning, the time elapsed since cleaning, and the number of cleanings, and a maintenance content judgment result, which is correct answer data (step S31). The data acquisition unit 24 stores the acquired data set including the measurement value data, the time elapsed since cleaning, the time elapsed since cleaning, and the number of cleanings, and the maintenance content judgment result in the training data storage unit 29.

[0057] Although the data set including the measurement value data, the time elapsed since cleaning, the time elapsed since cleaning, and the number of cleanings, and the maintenance content judgment result are acquired simultaneously, it is sufficient if the data set including the measurement value data, the time elapsed since cleaning, the time elapsed since cleaning, and the number of cleanings, and the maintenance content judgment result can be input in association with each other, and the data set including the measurement value data, the time elapsed since cleaning, the time elapsed since cleaning, and the number of cleanings, and the maintenance content judgment result may be acquired at different times.

[0058] Thereafter, the model generation unit 30 learns the maintenance timing by so-called supervised learning according to the learning data created based on a combination of a data set including the measurement value data, the time elapsed since cleaning, the time elapsed since cleaning, and the number of cleanings, and the maintenance content judgment results, and generates a maintenance timing inference model, which is a learned model (step S32).

[0059] Then, the trained model storage unit 25 stores the trained model generated by the model generation unit 30 (step S33). This completes the trained model generation process.

[0060] <How to determine when maintenance is required> 7 is a flowchart showing an example of the procedure for determining the timing of maintenance. First, the data acquisition unit 24 acquires a data set including measurement value data, time elapsed since cleaning, time elapsed since cleaning, and the number of cleanings (step S51).

[0061] Next, the inference unit 26 inputs a data set including the measurement value data, the elapsed time since cleaning, the elapsed time since cleaning, and the number of cleanings into the maintenance timing inference model, which is a learned model stored in the learned model memory unit 25, and obtains the maintenance timing (step S52).

[0062] Thereafter, the inference unit 26 outputs the maintenance timing obtained from the trained model to the maintenance processing unit 27 (step S53).

[0063] The maintenance processing unit 27 then uses the output maintenance timing to perform maintenance on the instrumentation sensor 11 if necessary (step S54). That is, the maintenance processing unit 27 performs processing according to the output maintenance timing. Specifically, if the result obtained is that maintenance is not required, the change in the measurement value data by the instrumentation sensor 11 is a normal measurement value change, so no special processing is performed and the current state is continued. If the result obtained is that cleaning is required, the maintenance processing unit 27 outputs cleaning alarm information to the display unit 28 and outputs a control signal to the cleaning machine 12 to instruct cleaning. The cleaning machine 12 cleans the instrumentation sensor 11 in accordance with the control signal. After cleaning of the instrumentation sensor 11 is completed, the instrumentation sensor 11 outputs a cleaning completion signal. When the data acquisition unit 24 receives the cleaning completion signal, it starts time measurement by the cleaning determination timer 21.

[0064] If the result is cleaning, the maintenance processing unit 27 outputs cleaning alarm information to the display unit 28. When the operator confirms the cleaning alarm information, he or she cleans the instrumentation sensor 11 while stopping the operation of the water treatment device 101. After completing cleaning of the instrumentation sensor 11, the operator outputs a cleaning completion signal from the information processing terminal. When the data acquisition unit 24 receives the cleaning completion signal, it starts timing on the re-cleaning determination timer 22. Note that if the cleaning alarm information includes the number of cleanings, the operator can perform maintenance taking into account the possibility that the instrumentation sensor 11 is abnormal. As a result, the instrumentation sensor 11 is cleaned when activated sludge adheres to it, and the instrumentation sensor 11 is cleaned before the activated sludge adhering to the instrumentation sensor 11 significantly affects the activated sludge concentration measurement value measured by the instrumentation sensor 11. This completes the process.

[0065] In the first embodiment, a case where supervised learning is applied to the learning algorithm used by the model generation unit 30 has been described, but the present invention is not limited to this. It is also possible to apply a learning algorithm other than supervised learning.

[0066] The model generation unit 30 may also learn the maintenance timing according to learning data created for multiple instrumentation sensors 11. The model generation unit 30 may acquire learning data from multiple instrumentation sensors 11 used in the same area, or may learn the maintenance timing using learning data collected from multiple instrumentation sensors 11 operating independently in different areas. It is also possible to add or remove instrumentation sensors 11 that collect learning data from the target during the process. Furthermore, a learning device that has learned the maintenance timing for a certain instrumentation sensor 11 may be applied to another instrumentation sensor 11, and the maintenance timing for the other instrumentation sensor 11 may be re-learned and updated.

[0067] Furthermore, the learning algorithm used in the model generation unit 30 can be deep learning, which learns to extract the features themselves, or machine learning can be performed according to other known methods, such as genetic programming, functional logic programming, or support vector machines.

[0068] Furthermore, in the example of FIG. 3, the maintenance timing determination device 20 is configured to include a processing unit that generates a trained model, but the processing unit that generates the trained model may be configured as a learning device separate from the maintenance timing determination device 20. That is, the learning device may include the data acquisition unit 24, the training data storage unit 29, and the model generation unit 30 of FIG. 3 and be capable of communicating with the maintenance timing determination device 20. In this case, the learning device may exist on a cloud server. Furthermore, the maintenance timing determination device 20 may also exist on the cloud server.

[0069] In the above example, a data set including measurement value data, time elapsed since cleaning, time elapsed since cleaning, and cleaning count is input into the maintenance timing inference model to determine the timing of cleaning or maintenance. As explained in the flowchart of Figure 5, if an abnormal change in measurement value occurs a predetermined number of times or more during the re-cleaning determination period T2, it is considered that an event that cannot be resolved by cleaning has occurred in the instrumentation sensor 11. In other words, if the number of cleanings counted by the cleaning count counter 23 is a predetermined number or more, it can be considered that an abnormality has occurred in the instrumentation sensor 11.

[0070] Therefore, the maintenance content judgment result corresponding to the data set in the learning data when the number of cleanings during the cleaning determination period T1 is "1" is set to "cleaning," and the maintenance content judgment result corresponding to the data set when the number of cleanings during the re-cleaning determination period T2 is "repair." By generating a maintenance timing inference model using this learning data, it becomes possible to infer whether maintenance is not required, whether cleaning is required, or whether cleaning and repair are required based on the measurement value data, the time elapsed since cleaning, the time elapsed since cleaning, and the number of cleanings. In this case, the maintenance processing unit 27, for example, outputs cleaning alarm information when the number of cleanings is "1," and outputs abnormality alarm information indicating the occurrence of an abnormality in the instrumentation sensor 11 when the number of cleanings is "2" or more.

[0071] Furthermore, in the above explanation, the data necessary to determine the timing of maintenance includes the measurement value data from the instrumentation sensor 11, the time elapsed since cleaning which is the value measured by the cleaning determination timer 21, the time elapsed since cleaning which is the value measured by the re-cleaning determination timer 22, and the number of cleanings counted by the cleaning count counting unit 23, but it may also include data other than these.

[0072] In the above description, the cleaning determination period T1 is a fixed value, but it is also possible to change the cleaning determination period T1. Therefore, the following description will be given as an example of determining the cleaning determination period T1 by machine learning.

[0073] Fig. 8 is a diagram schematically illustrating another example of the configuration of the maintenance timing determination device according to embodiment 1. Note that the same components as those in Fig. 3 are given the same reference numerals and their description will be omitted, and only the parts that differ from Fig. 3 will be described.

[0074] The maintenance timing determination device 20A shown in FIG.

[0075] The abnormality detection timer 31 starts timing when it detects an abnormal change in the measurement value, and measures the abnormality detection interval, which is the time until the next abnormal change in the measurement value is detected. When the abnormality detection timer 31 detects the next abnormal change in the measurement value, it resets the timer and then starts timing again. The abnormality detection timer 31 outputs the measured abnormality detection interval to the data acquisition unit 24. Since abnormal changes in the measurement value occur due to the adhesion of dirt to the instrumentation sensor 11, the adhesion of dirt to the instrumentation sensor 11 corresponds to an abnormal state of the instrumentation sensor 11.

[0076] When generating a trained model for inferring the cleaning determination period T1, the data acquisition unit 24 acquires, as training data, a data set including the elapsed time since cleaning and the anomaly detection interval, and cleaning determination period judgment result information, which is correct answer data. The data acquisition unit 24 stores the acquired training data in the training data storage unit 29. This accumulates the training data. The training data acquired for inferring the cleaning determination period T1 corresponds to the training data for the cleaning determination period T1.

[0077] The time elapsed since cleaning is data measured by the re-cleaning determination timer 22, and indicates the time elapsed since cleaning was completed. It is believed that the way in which dirt adheres to the instrumentation sensor 11 differs depending on the time that has passed since cleaning was completed. In other words, immediately after cleaning is completed, dirt is unlikely to adhere to the instrumentation sensor 11, but as time passes, dirt that cannot be completely removed by cleaning with the cleaning machine 12 gradually accumulates and becomes more likely to adhere. This is thought to be due to a change in the quality of the dirt. For this reason, the time elapsed since cleaning is thought to indicate the ease with which dirt adheres.

[0078] As described above, the abnormality detection interval is the time from the time when an abnormal change in measurement value is detected to the time when the next abnormal change in measurement value is detected. Since the longer the time that has passed since the completion of cleaning, the easier it is for dirt to adhere, it is thought that the abnormality detection interval also changes with the passage of time from the completion of cleaning. In one example, it is thought that the abnormality detection interval gradually becomes shorter as time passes from immediately after cleaning.

[0079] The cleaning determination period determination result information is information indicating the cleaning determination period T1 determined by a skilled worker based on the degree of dirt, i.e., the degree of dirt adhesion to the detection area 111, when cleaning the instrumentation sensor 11. Comparing washing and cleaning, cleaning requires more human labor, so it is preferable to remove dirt by washing as much as possible. Therefore, the cleaning determination period T1 is set long so that the cleaning trigger is less likely to be activated. On the other hand, as time passes after cleaning is completed, dirt that cannot be completely removed by cleaning with the cleaning device 12 becomes hardened.

[0080] Now, consider a case where processing is performed during a certain cleaning determination period T1. When the maintenance method shown in FIG. 5 is performed during the cleaning determination period T1, if cleaning is completed at a certain time and the cleaning determination period T1 has elapsed without detecting an abnormal change in the measurement value, this corresponds to No in step S14, and the process does not proceed to cleaning. In other words, cleaning is performed when an abnormal change in the measurement value is detected after the cleaning determination period T1 has elapsed. When this process is repeated, dirt that was not completely removed by cleaning gradually accumulates and adheres to the detection area 111. Then, an abnormal change in the measurement value is detected after a time elapsed since cleaning that is shorter than the cleaning determination period T1, and cleaning is performed.

[0081] During cleaning, an operator observes and cleans the detection area 111 of the instrumentation sensor 11, and can determine whether the dirt can be removed by cleaning or washing depending on the degree of dirt on the detection area 111 at this time. In the example described above, the detection area 111 includes a light emission area that emits light and a light detection area that detects scattered light. If dirt adheres to the light emission area and the light detection area of ​​the detection area 111, it is highly likely that the detection of the activated sludge concentration will be affected, but if dirt adheres to areas other than the light emission area and the light detection area, it is considered unlikely that the detection of the activated sludge concentration will be affected.

[0082] Furthermore, even when dirt is attached to the light emission area and the light detection area, it is possible to determine whether the dirt can be removed by washing or can only be removed by cleaning. If the dirt can be removed by washing, the current cleaning determination period T1 is long, and the dirt that can be removed by washing has been determined to be dirt that must be cleaned. In other words, it can be determined that the cleaning determination period T1 is too long for the current state of the instrumentation sensor 11. A skilled worker can determine the cleaning determination period T1 based on the degree of dirt so that the attachment of such dirt does not fall under "cleaning" but is determined to be "cleaning." On the other hand, if the dirt can only be removed by cleaning, the skilled worker can determine that the cleaning determination period T1 matches the current state of the instrumentation sensor 11.

[0083] As described above, the cleaning determination period T1 determined by the skilled worker based on the degree of dirt on the detection area 111 of the instrumentation sensor 11 becomes the cleaning determination period determination result information.

[0084] In addition to the function of generating a model for inferring the maintenance timing described above, the model generation unit 30 also has a function of generating a cleaning determination period inference model, which is a trained model for inferring the cleaning determination period T1. Specifically, the model generation unit 30 learns the cleaning determination period T1 based on training data created based on a combination of a data set including the elapsed time since cleaning and the abnormality detection interval acquired from the training data storage unit 29 and the cleaning determination period judgment result information, which is the correct answer data. That is, the model generation unit 30 generates a trained model that infers the optimal cleaning determination period T1 from the data set including the elapsed time since cleaning and the abnormality detection interval of the maintenance timing determination device 20A and the cleaning determination period judgment result information. Here, the training data is data in which the data set including the elapsed time since cleaning and the abnormality detection interval and the cleaning determination period judgment result information are associated with each other.

[0085] The learning algorithm used by the model generation unit 30 can be a known algorithm such as supervised learning. As an example, similar to the case of the model that infers the maintenance timing described above, the cleaning determination period T1 is learned by so-called supervised learning according to a neural network model. By performing the above-described learning, the model generation unit 30 generates a cleaning determination period inference model, which is a learned model, and outputs it to the learned model storage unit 25.

[0086] When the cleaning determination period T1 is inferred using the cleaning determination period inference model, the data acquisition unit 24 acquires a data set including the elapsed time since cleaning and the abnormality detection interval.

[0087] In addition to the function of inferring the maintenance timing described above, the inference unit 26 also has a function of inferring the cleaning determination period T1 using a cleaning determination period inference model. Specifically, the inference unit 26 infers the cleaning determination period T1 obtained using the cleaning determination period inference model. That is, by inputting a data set including the elapsed time since cleaning and the abnormality detection interval acquired by the data acquisition unit 24 into this cleaning determination period inference model, it is possible to output the cleaning determination period T1 inferred from the data set including the elapsed time since cleaning and the abnormality detection interval.

[0088] In this example, the cleaning judgment period T1 is output using a cleaning judgment period inference model learned by the model generation unit 30 of the maintenance timing judgment device 20A, but it is also possible to obtain a cleaning judgment period inference model from an external source, such as another maintenance timing judgment device 20A, and output the cleaning judgment period T1 based on this cleaning judgment period inference model.

[0089] The model generation unit 30 may also learn the cleaning determination period T1 using learning data acquired from multiple maintenance timing determination devices 20A. The model generation unit 30 may acquire learning data from multiple maintenance timing determination devices 20A used in the same area, or may learn the cleaning determination period T1 using learning data collected from multiple maintenance timing determination devices 20A operating independently in different areas. It is also possible to add or remove maintenance timing determination devices 20A that collect learning data from the target during the process. Furthermore, a learning device that has learned the cleaning determination period T1 for a certain maintenance timing determination device 20A may be applied to another maintenance timing determination device 20A, and the cleaning determination period T1 for the other maintenance timing determination device 20A may be re-learned and updated.

[0090] Furthermore, the method for generating the cleaning determination period inference model is the same as that shown in Fig. 6, and the method for inferring the cleaning determination period T1 is the same as that shown in Fig. 7, so their explanations are omitted. However, in the process corresponding to step S54 in Fig. 7, the cleaning determination period T1 obtained by the cleaning determination period inference model is used in the maintenance timing inference model that infers the maintenance timing of the inference unit 26. This results in a cleaning determination period T1 of a length corresponding to the elapsed time since cleaning and the abnormality detection interval, making it possible to prevent a situation that should be determined as cleaned from being erroneously determined as cleaned.

[0091] As described above, in the first embodiment, the timing of maintenance is estimated based on the measurement data, the time elapsed since cleaning, the time elapsed since cleaning, and the number of cleanings. This enables cleaning or cleaning to be performed when dirt adheres to the detection area 111 of the instrumentation sensor 11 and depending on the quality of the dirt. Conventionally, cleaning using the cleaning machine 12 is performed at predetermined intervals, such as every hour, and cleaning is performed at predetermined intervals, such as every month. Therefore, if dirt adheres before the predetermined period, the dirt becomes difficult to remove even with regular cleaning, and the instrumentation sensor 11 must be cleaned earlier than the predetermined period. However, in the first embodiment, the instrumentation sensor 11 can be cleaned when dirt adheres. As a result, the period during which dirt solidifies before cleaning is necessary can be extended. In other words, the cleaning interval of the instrumentation sensor 11 can be extended. As a result, when comparing the number of cleanings in a predetermined period, the maintenance timing determination device 20, 20A according to the first embodiment can extend the cleaning interval, thereby reducing the number of cleanings in a predetermined period compared to conventional methods. Although the operation of the water treatment device 101 must be stopped during cleaning, the number of cleanings required in a given period, i.e., the number of interruptions, is reduced compared to conventional methods, so that the operation suspension period can be shortened. Furthermore, because the operation suspension period is shortened, it is possible to reduce the labor required for maintenance compared to conventional methods.

[0092] Furthermore, if the conventional cleaning interval is set short, cleaning may be performed even when activated sludge is not attached to the instrumentation sensor 11. In this case, cleaning is wasted and cleaning costs are incurred. However, in the first embodiment, cleaning is performed when activated sludge adheres, so there is no more unnecessary cleaning than in the past, and it is possible to reduce cleaning costs.

[0093] Embodiment 2 In embodiment 1, the timing of maintenance was determined using the measurement value of the instrumentation sensor 11, but in embodiment 2, a maintenance timing determination device 20 and a maintenance timing determination method are described that determine the timing of maintenance using an image including the detection area 111 of the instrumentation sensor 11.

[0094] The maintenance timing determination device according to the second embodiment has the same configuration as that described in the first embodiment. Below, a processing unit having a different function from that in the first embodiment will be described using the maintenance timing determination device 20 in Fig. 3 as an example.

[0095] When inferring the timing of maintenance, the data acquisition unit 24 includes image data including the detection area 111 of the instrumentation sensor 11, the time elapsed since washing, the time elapsed since cleaning, and the number of cleanings as data necessary for determining the timing of maintenance. That is, in the second embodiment, instead of the measurement value data of the instrumentation sensor 11 in the first embodiment, image data of the detection area 111 is acquired.

[0096] 9 is a diagram schematically illustrating an example of the configuration of an instrumentation device connected to a maintenance timing determination device according to embodiment 2. The instrumentation sensor 11 has a housing 110, a detector 112, and a camera 113, which is an imaging unit. The detector 112 and the camera 113 are disposed inside the housing 110. Of the surfaces constituting the outer periphery of the housing 110, the portion that is placed in the flow of the water to be treated inside the pipe 15 when the instrumentation sensor 11 is installed in the pipe 15 is referred to as the detection surface 110a.

[0097] The detector 112 measures the concentration of activated sludge in the water to be treated. The detector 112 is disposed inside the housing 110 so that one end thereof contacts the detection surface 110a of the housing 110. The area including the area where the detector 112 and the detection surface 110a of the housing 110 contact each other becomes the detection area 111. When the instrumentation sensor 11 is a scattered light type sludge concentration meter, the detector 112 consists of a light emitting unit that emits light and a light receiving unit that receives scattered light. The area including the detection area 111 becomes the cleaning area that is cleaned by the cleaning machine 12. The results detected by the detector 112 are output to the maintenance timing determination device 20.

[0098] The camera 113 captures an image of an area of ​​the detection surface 110a of the housing 110, including the detection area 111. The area captured by the camera 113 is referred to as the imaging area. Since the camera 113 captures an image of the imaging area from inside the housing 110, the components of the detector 112 on the detection surface 110a side are made of a transparent material. Furthermore, since the camera 113 captures an image of the side of the detection area 111 that comes into contact with the water to be treated, at least the detection area 111 of the detection surface 110a is made of a transparent material. Image data of the imaging area captured by the camera 113 is output to the maintenance timing determination device 20.

[0099] The image data is image data of an imaging area including the detection area 111 captured by the camera 113, and therefore includes the state of dirt adhering to the detection area 111.

[0100] The trained model storage unit 25 stores a maintenance timing inference model, which is a trained model that infers maintenance timing from image data, the time elapsed since cleaning, the time elapsed since cleaning, and the number of cleanings.

[0101] The inference unit 26 can output the maintenance timing inferred from the image data, the time elapsed since cleaning, the time elapsed since cleaning, and the number of cleanings by inputting the image data, the time elapsed since cleaning, the time elapsed since cleaning, and the number of cleanings acquired by the data acquisition unit 24 into the maintenance timing inference model. The inference unit 26 outputs the inference result to the maintenance processing unit 27.

[0102] When generating a trained model, the data acquisition unit 24 acquires data necessary for training the maintenance timing inference model and stores it in the training data storage unit 29. The data necessary for training the maintenance timing inference model is training data that associates a data set including image data, the time elapsed since cleaning, the time elapsed since cleaning, and the number of cleanings with the maintenance content judgment result, which is the correct answer data at that time. Note that, in the second embodiment, as in the first embodiment, a maintenance content judgment result, which is the judgment result made by a skilled worker, is used when cleaning the instrumentation sensor 11. The judge determines the maintenance content judgment result by taking into account whether the time elapsed since cleaning is within a predetermined cleaning judgment period T1, whether the time elapsed since cleaning is within a predetermined re-cleaning judgment period T2, and the number of cleanings. However, in order to infer whether maintenance is not required or whether cleaning and cleaning are required using the image data, image data may be acquired at various times, and the skilled worker may judge the degree of dirtiness of the instrumentation sensor 11 at the time the image data is acquired.

[0103] The learning data storage unit 29 stores a plurality of learning data sets that correspond to a data set including image data, time elapsed since washing, time elapsed since cleaning, and number of cleanings, and correct answer data including the judgment result at that time.

[0104] The model generation unit 30 learns the maintenance timing based on learning data created based on a combination of a data set including the image data, the time elapsed since cleaning, the time elapsed since cleaning, and the number of cleanings acquired by the data acquisition unit 24, i.e., stored in the learning data storage unit 29, and supervised data including the maintenance content judgment result at that time. That is, a maintenance timing inference model is generated that infers the optimal maintenance timing from the data set including the image data of the instrumentation sensor 11, the time elapsed since cleaning, the time elapsed since cleaning, and the number of cleanings, and the supervised data including the maintenance content judgment result at that time.

[0105] When performing processing using image data, the model generation unit 30 may, for example, use differential image data between the image data and reference image data, which is image data when the instrumentation sensor 11 is in an unused state. The model generation unit 30 acquires characteristic information about the dirt, including its position, size, and thickness, from the differential image data. The thickness can be inferred from the shading of the differential image data. The model generation unit 30 then learns the correlation between the characteristic information about the dirt, including its position, size, and thickness, the time elapsed since cleaning, and the maintenance content judgment result.

[0106] In addition, when performing inference, the inference unit 26 also acquires characteristic information of the dirt using differential image data between the image data and the reference image data, and by inputting the acquired characteristic information, the elapsed time since cleaning, the elapsed time since cleaning, and the number of cleanings into the model, it is possible to obtain the maintenance timing.

[0107] In the above explanation, the data necessary to determine the timing of maintenance includes image data, the time elapsed since cleaning which is the value measured by the cleaning determination timer 21, the time elapsed since cleaning which is the value measured by the re-cleaning determination timer 22, and the number of cleanings counted by the cleaning count count unit 23, but it may also include data other than these.

[0108] The method for generating a trained model and the method for determining maintenance timing in the second embodiment are the same as those in the first embodiment, except for the type of data used, and therefore the description thereof will be omitted. Also in the second embodiment, as described in the first embodiment, a cleaning determination period inference model that infers the cleaning determination period T1 from a data set including the time elapsed since cleaning and the anomaly detection interval may be generated, and the cleaning determination period T1 obtained by inputting the data set including the time elapsed since cleaning and the anomaly detection interval into this cleaning determination period inference model may be used in the maintenance timing inference model.

[0109] In the second embodiment, image data of an imaging area including the detection area 111 of the instrumentation sensor 11 captured by the camera 113, the time elapsed since cleaning, the time elapsed since cleaning, and the number of cleanings are input into a model to acquire the maintenance timing. This has the effect of detecting the occurrence of dirt from the image data of the detection area 111 and performing appropriate maintenance according to the degree of dirt. Furthermore, when determining the maintenance timing from the measurement value of the instrumentation sensor 11 as in the first embodiment, it can be difficult to distinguish whether a change in the measurement value is due to a change in the concentration of activated sludge in the water to be treated, due to the adhesion of dirt, or due to a malfunction of the instrumentation sensor 11. However, in the second embodiment, the adhesion of dirt is detected using image data, and therefore the accuracy of the determination can be improved compared to the first embodiment.

[0110] In the above description, the cleaning determination timer 21 and the re-cleaning determination timer 22 are provided inside the maintenance timing determination device 20, but they may also be provided outside the maintenance timing determination device 20.

[0111] Next, the hardware configuration of the maintenance timing determination device 20, 20A according to the first and second embodiments will be described. The maintenance timing determination device 20, 20A according to the first and second embodiments functions as the maintenance timing determination device 20, 20A by executing a computer program on the computer system, which is a computer program that describes the processing of the maintenance timing determination device 20, 20A. FIG. 10 is a block diagram showing an example of the configuration of a computer system that realizes the maintenance timing determination device according to the first and second embodiments. As shown in FIG. 10, this computer system includes a control unit 81, an input unit 82, a storage unit 83, a display unit 84, a communication unit 85, and an output unit 86, which are connected via a system bus 87.

[0112] In FIG. 10 , the control unit 81 is a processor such as a CPU (Central Processing Unit) that executes a program describing the processing of the maintenance timing determination device 20 or 20A according to the first or second embodiment. Note that a portion of the control unit 81 may be implemented by dedicated hardware such as a GPU (Graphics Processing Unit) or an FPGA (Field-Programmable Gate Array). The input unit 82 includes a keyboard, a mouse, and the like, and is used by a user of the computer system to input various pieces of information. The memory unit 83 includes various types of memory, such as RAM (Random Access Memory) and ROM (Read Only Memory), and a storage device, such as a hard disk, and stores programs to be executed by the control unit 81, necessary data obtained during processing, and the like. The memory unit 83 is also used as a temporary storage area for programs. The display unit 84 includes a display, a liquid crystal display (LCD), and the like, and displays various screens to the user of the computer system. The communication unit 85 is a receiver and transmitter that perform communication processing. The output unit 86 includes a printer, a speaker, and the like. Note that FIG. 10 is just an example, and the configuration of the computer system is not limited to the example of FIG.

[0113] Here, an example of the operation of the computer system until the above-mentioned program is ready to be executed will be described. In the computer system having the above-mentioned configuration, for example, a computer program is installed in the storage unit 83 from a CD-ROM or DVD-ROM inserted in a CD (Compact Disc)-ROM drive or DVD (Digital Versatile Disc)-ROM drive (not shown). Then, when the program is executed, the program read from the storage unit 83 is stored in the main storage area of ​​the storage unit 83. In this state, the control unit 81 executes the processing as the maintenance timing determination device 20 or 20A according to the first or second embodiment in accordance with the program stored in the storage unit 83.

[0114] In the above description, a program describing the processing in the maintenance timing determination device 20, 20A is provided using a CD-ROM or DVD-ROM as a recording medium, but this is not limited to this. Depending on the configuration of the computer system, the capacity of the program to be provided, etc., it is also possible to use a program provided via a transmission medium such as the Internet via the communication unit 85.

[0115] In one example, this computer program causes a computer system to execute the processing procedure shown in FIG. 6 or the processing procedure shown in FIG.

[0116] The cleaning determination timer 21, re-cleaning determination timer 22, cleaning count count unit 23, inference unit 26, maintenance processing unit 27, model generation unit 30, and abnormality detection timer 31 shown in FIGS. 3 and 8 are realized by the control unit 81 executing a computer program stored in the memory unit 83. The memory unit 83 is also used to realize the cleaning determination timer 21, re-cleaning determination timer 22, cleaning count count unit 23, inference unit 26, maintenance processing unit 27, model generation unit 30, and abnormality detection timer 31 shown in FIGS. 3 and 8. The data acquisition unit 24 shown in FIGS. 3 and 8 is realized by the communication unit 85. The trained model storage unit 25 and the training data storage unit 29 shown in FIGS. 3 and 8 are realized by the storage unit 83. The display unit 28 shown in FIGS. 3 and 8 is realized by the display unit 84 shown in FIG. 10. The maintenance timing determination device 20, 20A may be realized by multiple computer systems. In one example, the maintenance timing determination device 20, 20A may be realized by a cloud computer system.

[0117] The configurations shown in the above embodiments are merely examples, and may be combined with other known technologies, or different embodiments may be combined with each other. It is also possible to omit or modify parts of the configurations as long as they do not deviate from the gist of the invention. [Explanation of symbols]

[0118] 1 influent piping, 2 anoxic tank, 3 aerobic tank, 4 final sedimentation tank, 5 nitrification liquid circulation pump, 6 aeration device, 7 blower, 8 submersible agitator, 9 sludge extraction pump, 11, 11a, 11b instrumentation sensor, 12, 12a, 12b cleaning machine, 15 piping, 20, 20A maintenance timing determination device, 21 cleaning determination timer, 22 re-cleaning determination timer, 23 cleaning count unit, 24 data acquisition unit, 25 trained model memory unit, 26 inference unit, 27 maintenance processing unit, 28 display unit, 29 learning data memory unit, 30 model generation unit, 31 abnormality detection timer, 100 water treatment system, 101 water treatment device, 110 housing, 110a detection surface, 111 detection area, 112 detector, 113 camera, 121 injection nozzle, 122 valve, 151 Through hole.

Claims

1. A maintenance timing determination device that is provided in a water treatment device and determines the timing of maintenance of an instrumentation sensor that measures pollutants in water to be treated, a data acquisition unit that acquires measurement data measured by the instrumentation sensor at one or more times, a post-cleaning time that is the time that has elapsed since cleaning of the instrumentation sensor by a cleaning machine was completed, a post-cleaning time that is the time that has elapsed since cleaning performed by removing the instrumentation sensor was completed, and a cleaning count that is the number of times that cleaning was determined; an inference unit that outputs the maintenance timing from the measurement value data, the elapsed time since cleaning, the elapsed time since cleaning, and the number of cleanings inputted from the data acquisition unit, using a maintenance timing inference model that is a trained model for inferring the timing of maintenance as either no maintenance required or cleaning and cleaning based on the measurement value data, the elapsed time since cleaning, the elapsed time since cleaning, and the number of cleanings; a maintenance processing unit that performs processing in accordance with the maintenance timing output from the inference unit; A maintenance timing determination device comprising:

2. The data acquisition unit further has a function of acquiring learning data including the measurement value data, the time elapsed since cleaning, the time elapsed since cleaning, and the number of cleanings, and a maintenance content determination result that is a result of determining whether the maintenance is unnecessary, the cleaning, or the cleaning based on the state of the instrumentation sensor at that time, The maintenance timing determination device according to claim 1, further comprising a model generation unit that uses the learning data to generate the maintenance timing inference model for inferring the maintenance timing from the measurement value data, the elapsed time since cleaning, the elapsed time since cleaning, and the number of cleanings.

3. A maintenance timing determination device that is provided in a water treatment device and determines the timing of maintenance of an instrumentation sensor that measures pollutants in water to be treated, a data acquisition unit that acquires image data of an area including a detection area of ​​the instrumentation sensor that detects the contaminants, a post-cleaning time that is the time that has elapsed since the instrumentation sensor was cleaned by a cleaning machine, a post-cleaning time that is the time that has elapsed since the instrumentation sensor was removed and cleaning was completed, and a cleaning count that is the number of times that cleaning was determined; an inference unit that outputs the maintenance timing from the image data, the elapsed time since cleaning, the elapsed time since cleaning, and the number of cleanings input from the data acquisition unit, using a maintenance timing inference model that is a trained model for inferring the timing of maintenance as either no maintenance required or cleaning and cleaning based on the image data, the elapsed time since cleaning, the elapsed time since cleaning, and the number of cleanings; a maintenance processing unit that performs processing in accordance with the maintenance timing output from the inference unit; A maintenance timing determination device comprising:

4. The data acquisition unit further has a function of acquiring learning data including the image data, the time elapsed since cleaning, the time elapsed since cleaning, and the number of cleanings, and a maintenance content determination result that is a result of determining whether the maintenance is unnecessary, the cleaning, or the cleaning based on the state of the instrumentation sensor at that time, The maintenance timing determination device according to claim 3, further comprising a model generation unit that uses the learning data to generate the maintenance timing inference model for inferring the maintenance timing from the image data, the time elapsed since cleaning, the time elapsed since cleaning, and the number of cleanings.

5. the inference unit uses a cleaning determination period, which is a period during which it is determined that the cleaning should be performed instead of the cleaning if an abnormality in the state of the instrumentation sensor is detected after the cleaning of the instrumentation sensor is completed, when inferring the maintenance timing; The data acquisition unit further has a function of acquiring learning data for a cleaning determination period, including a data set including the time elapsed since cleaning and an abnormality detection interval, which is the time from the detection of an abnormality in the state of the instrumentation sensor to the detection of the next abnormality in the state, and cleaning determination period determination result information, which is the cleaning determination period determined from the degree of dirtiness of the instrumentation sensor; The maintenance timing determination device described in claim 2 or 4, characterized in that the model generation unit further has a function of generating a cleaning determination period inference model, which is a trained model for inferring the cleaning determination period from the elapsed time since cleaning and the cleaning determination period determination result information, using learning data for the cleaning determination period.

6. A maintenance timing determination method in which a control device determines the timing of maintenance of an instrumentation sensor provided in a water treatment device that measures pollutants in water to be treated, comprising: a data acquisition process in which the control device acquires measurement data measured by the instrumentation sensor at one or more times, a post-cleaning elapsed time which is the elapsed time since cleaning of the instrumentation sensor by a cleaning machine was completed, a post-cleaning elapsed time which is the elapsed time since cleaning performed by removing the instrumentation sensor was completed, and a cleaning count which is the number of times cleaning was determined; an inference step in which the control device outputs the maintenance timing from the measurement value data, the elapsed time since cleaning, the elapsed time since cleaning, and the number of cleanings acquired and input in the data acquisition step, using a trained model for inferring whether the maintenance timing is no maintenance, cleaning, or cleaning from the measurement value data, the elapsed time since cleaning, the elapsed time since cleaning, and the number of cleanings; a maintenance processing step in which the control device performs processing according to the maintenance timing output in the inference step; A maintenance timing determination method comprising:

7. A maintenance timing determination method in which a control device determines the timing of maintenance of an instrumentation sensor provided in a water treatment device that measures pollutants in water to be treated, comprising: a data acquisition process in which the control device acquires image data of an imaging area including a detection area for detecting the pollutants of the instrumentation sensor, a post-cleaning elapsed time which is the elapsed time since the instrumentation sensor was cleaned by a cleaning machine, a post-cleaning elapsed time which is the elapsed time since the instrumentation sensor was removed and cleaning was completed, and a cleaning count which is the number of times cleaning was determined; an inference step in which the control device outputs the maintenance timing from the image data, the elapsed time since cleaning, the elapsed time since cleaning, and the number of cleanings acquired and input in the data acquisition step, using a maintenance timing inference model that is a trained model for inferring whether the maintenance timing is no maintenance, cleaning, or cleaning from the image data, the elapsed time since cleaning, the elapsed time since cleaning, and the number of cleanings; a maintenance processing step in which the control device performs processing according to the maintenance timing output in the inference step; A maintenance timing determination method comprising:

8. A maintenance timing determination program that causes a computer system to function as a maintenance timing determination device that is provided in a water treatment device and determines the timing of maintenance of an instrumentation sensor that measures pollutants in water to be treated, The computer system includes: a data acquisition process for acquiring measurement data measured by the instrumentation sensor at one or more times, a post-cleaning time which is the time elapsed since cleaning of the instrumentation sensor by a cleaning machine was completed, a post-cleaning time which is the time elapsed since cleaning performed by removing the instrumentation sensor was completed, and a cleaning count which is the number of times cleaning was determined; an inference process of outputting the maintenance timing from the measurement value data, the elapsed time since cleaning, the elapsed time since cleaning, and the number of cleanings acquired and input in the data acquisition process, using a maintenance timing inference model which is a trained model for inferring the timing of maintenance as either no maintenance, cleaning, or cleaning from the measurement value data, the elapsed time since cleaning, the elapsed time since cleaning, and the number of cleanings; a maintenance processing step of performing processing according to the maintenance timing outputted in the inference step; A maintenance timing determination program that executes the above steps.

9. A maintenance timing determination program that causes a computer system to function as a maintenance timing determination device that is provided in a water treatment device and determines the timing of maintenance of an instrumentation sensor that measures pollutants in water to be treated, The computer system includes: a data acquisition process for acquiring image data of an area including a detection area of ​​the instrumentation sensor where the pollutants are detected, a post-cleaning time which is the time elapsed since the instrumentation sensor was cleaned by a cleaning machine, a post-cleaning time which is the time elapsed since the instrumentation sensor was removed and cleaning was completed, and a cleaning count which is the number of times cleaning was determined; an inference process of outputting the maintenance timing from the image data, the elapsed time since cleaning, the elapsed time since cleaning, and the number of cleanings acquired and input in the data acquisition process, using a maintenance timing inference model that is a trained model for inferring the timing of maintenance as either no maintenance, cleaning, or cleaning from the image data, the elapsed time since cleaning, the elapsed time since cleaning, and the number of cleanings; a maintenance processing step of performing processing according to the maintenance timing outputted in the inference step; A maintenance timing determination program that executes the above steps.

Images