Viscosity measurement method and viscosity measurement device

Abstract

To provide a viscosity measurement method that can correctly determine whether the viscosity of sludge in a chamber measured by a sludge viscosity measuring device is appropriate.SOLUTION: This is a viscosity measurement method for measuring the viscosity of activated sludge in a biological treatment tank equipped with a membrane separation device, and involves alternately repeating one cycle of sludge viscosity measurement, in which the sludge viscosity is measured over time for activated sludge filled in a chamber equipped with a vibration viscometer, and a reference viscosity measurement, in which a reference viscosity is measured over time for a cleaning solution filled in the chamber, to obtain multiple cycles of measurement data consisting of a set of sludge viscosity and reference viscosity obtained over time, and evaluating whether the data is normal or abnormal based on the fluctuation characteristics of the obtained multiple cycles of measurement data. A representative sludge viscosity of activated sludge is calculated and output from the measurement data evaluated as normal data.SELECTED DRAWING: Figure 3

Description

[Technical Field]

[0001] The present invention relates to a viscosity measurement method and a viscosity measurement device for measuring the viscosity of activated sludge filled in a biological treatment tank in which a membrane separation device is submerged. [Background technology]

[0002] In water treatment equipment that uses the membrane bioreactor, the optimum membrane filtration rate and the timing of membrane cleaning are determined based on the degree of fluctuation in the transmembrane pressure (TMP) of the membrane separation device, which is immersed in the biological treatment tank.

[0003] However, when the degree of membrane clogging is determined using the transmembrane pressure (TMP) as an index, the timing of the determination tends to be delayed, which can result in excessive aeration and unnecessary membrane cleaning, which can increase maintenance costs.

[0004] Therefore, Patent Document 1 proposes a sludge viscosity measuring device that can detect a clogged state of a membrane earlier than detecting a change in transmembrane pressure TMP. The sludge viscosity measuring device includes a chamber for storing a sludge sample, a vibration viscometer for measuring the viscosity of the sludge sample in the chamber, a sample supply system for supplying the sludge sample into the chamber, a cleaning liquid supply system for supplying a cleaning liquid into the chamber, a sample discharge system for discharging the sludge sample from the chamber, and a control device.

[0005] The control device repeatedly executes, in this order, a sludge supply process in which sample sludge is supplied into the chamber from the sample supply system; a viscosity measurement process in which the sludge sample is stored in the chamber and the viscosity of the sludge sample held in the chamber is measured using a vibration viscometer; a cleaning process in which cleaning liquid is supplied into the chamber from the cleaning liquid supply system and the sludge sample is discharged from the chamber to the sample discharge system; and a standby process in which the cleaning liquid supplied from the cleaning liquid supply system is stored in the chamber. [Prior art documents] [Patent documents]

[0006] [Patent Document 1] Patent Publication No. 2021-152522 Summary of the Invention [Problem to be solved by the invention]

[0007] However, when measuring sludge viscosity using the above-mentioned sludge viscosity measuring device, if foreign matter such as sediment or pebbles gets mixed in the sludge sample stored in the chamber, the sediment or pebbles may become entangled in the probe of the vibration viscometer, hindering the vibration of the probe and making it impossible to accurately determine the viscosity.

[0008] Furthermore, there is a risk that the sludge viscosity measured by the sludge viscosity measuring device will transiently fluctuate in accordance with changes in the operating state of the membrane separation device, making it impossible to measure an appropriate sludge viscosity.

[0009] If the amount of sludge extracted from the biological treatment tank were controlled based on such sludge viscosity, it would be difficult to maintain the activated sludge in an appropriate state, and it would be very difficult for an operator to determine whether the sludge viscosity measured by the sludge viscosity measuring device was an appropriate value.

[0010] In view of the above-mentioned problems, an object of the present invention is to provide a viscosity measurement method and a viscosity measurement device that can correctly determine whether the value measured by a sludge viscosity measurement device for sludge stored in a chamber is appropriate. [Means for solving the problem]

[0011] In order to achieve the above-mentioned object, the first characteristic configuration of the viscosity measurement method according to the present invention is a viscosity measurement method for measuring the viscosity of activated sludge filled in a biological treatment tank in which a membrane separation device is submerged, and the method comprises: a data acquisition process for acquiring multiple cycles of measurement data consisting of a set of the sludge viscosity and the reference viscosity acquired in a time series by alternately repeating one cycle of sludge viscosity measurement for measuring the sludge viscosity in a time series of the activated sludge filled in a chamber equipped with a vibration viscometer and a reference viscosity measurement for measuring a reference viscosity in a time series of the cleaning liquid filled in the chamber; an evaluation process for evaluating whether the data is normal or abnormal based on the fluctuation characteristics of the measurement data acquired in the multiple cycles in the data acquisition process; and a sludge viscosity output process for calculating and outputting a representative sludge viscosity of the activated sludge from the measurement data evaluated as normal data by the evaluation process.

[0012] In the data acquisition process, sludge viscosity measurements are performed in which activated sludge filled in a chamber is measured over time using a vibration viscometer to obtain multiple sludge viscosities in a time series, and reference viscosity measurements are performed in which cleaning liquid filled in the chamber is measured over time using a vibration viscometer to obtain multiple cycles of measurement data each consisting of a sludge viscosity and a reference viscosity. Based on the fluctuation characteristics of the measurement data over these multiple cycles, the evaluation process evaluates whether the data is normal or abnormal. In the sludge viscosity output process, a representative sludge viscosity of the activated sludge is calculated from the measurement data evaluated as normal and output.

[0013] The second characteristic configuration is that, in addition to the first characteristic configuration described above, the evaluation step includes a machine learning device that has performed machine learning in advance based on teacher data that has been labeled with at least the measurement data when the amount of filtered water in the membrane separation device fluctuates, the measurement data when foreign matter is attached to the vibration viscometer, and the measurement data when the membrane separation device has a constant amount of filtered water and no foreign matter is attached to the vibration viscometer, and the evaluation step is a step of evaluating the measurement data to be normal data if the label output when the measurement data acquired in the data acquisition step is input into the machine learning device indicates normal times.

[0014] Sludge viscosity measured over time using a vibration viscometer typically increases gradually from the start of measurement and then saturates at a value that represents the sludge's inherent viscosity. However, if foreign matter mixed in the sludge clings to the vibration viscometer's probe, the probe's vibration is suppressed, causing the value to continue to rise without saturating. Furthermore, the reference viscosity measured over time using a vibration viscometer for a cleaning solution typically shows a constant reference viscosity because the probe is cleaned with the cleaning solution and vibration is not suppressed. However, if foreign matter clinging to the probe cannot be removed during sludge viscosity measurement, the value will be higher than the reference viscosity. Furthermore, fluctuations in the operating conditions, such as the amount of filtrate in the membrane separation unit, can transiently fluctuate the viscosity of the activated sludge filled in the biological treatment tank, making it impossible to measure a stable sludge viscosity under steady conditions. For example, when the membrane is aerated and cleaned while membrane filtration is stopped, or when the membrane filtration rate decreases, the sludge viscosity appears to transiently decrease, while when the membrane filtration rate increases, the sludge viscosity appears to transiently increase, showing trends that differ from the steady state.

[0015] Therefore, teacher data is prepared that is labeled with at least the measurement data that transiently fluctuates when the amount of filtered water in the membrane separation device fluctuates, the abnormal measurement data when foreign matter adheres to the vibration viscometer, and the measurement data under normal conditions when the membrane separation device has a constant amount of filtered water and no foreign matter adheres to the vibration viscometer.A trained model is prepared that has been machine-learned so that a label corresponding to each measurement data is output from the machine learning device for each measurement data, and the measurement data is input into the trained model.This automatically determines which label the measurement data matches, and if the label indicates normal conditions, the measurement data is appropriately evaluated as normal data.

[0016] The third characteristic configuration is that, in addition to the first characteristic configuration described above, the evaluation process is a process of evaluating the measurement data as normal data when it is determined that there are no outliers in the measurement data using LOF or One-Class SVM learned based on the measurement data of multiple cycles under normal conditions acquired before the data acquisition process.

[0017] To perform sufficient learning using a machine learning device, it is necessary to prepare measurement data from various abnormal situations as training data, which makes the preparatory work extremely cumbersome. However, if learning can be performed using only measurement data from normal situations as semi-training data, the above-mentioned preparatory work becomes almost unnecessary. Therefore, it is possible to suitably use LOF (Local Outlier Factor), which detects density-based outliers, or One-Class SVM (Support Vector Machine), an outlier detection method that modifies the evaluation function of SVM.

[0018] The fourth characteristic configuration is that, in addition to any of the first to third characteristic configurations described above, the representative viscosity output in the sludge viscosity output process is the average value of the saturated values ​​of the sludge viscosity contained in the measurement data of the multiple cycles.

[0019] As described above, the sludge viscosity measured over time by a vibration viscometer usually gradually increases from the start of measurement and then saturates at a value that indicates the inherent viscosity of the sludge. This saturated value corresponds to the sludge viscosity. Therefore, an appropriate sludge viscosity can be obtained by calculating the average value of the saturated sludge viscosity values ​​contained in measurement data from multiple cycles as the representative viscosity.

[0020] The first characteristic configuration of the viscosity measuring device according to the present invention is that it is a viscosity measuring device for measuring the viscosity of activated sludge filled in a biological treatment tank in which a membrane separation device is submerged, and is equipped with a data acquisition unit that acquires measurement data for multiple cycles, each set consisting of the sludge viscosity and the reference viscosity, obtained in a time series by alternately repeating one cycle: a sludge viscosity measurement that measures the sludge viscosity of the activated sludge filled in a chamber equipped with a vibration viscometer, and a reference viscosity measurement that measures the reference viscosity of the cleaning liquid filled in the chamber in a time series; an evaluation unit that evaluates whether the data is normal or abnormal based on the fluctuation characteristics of the measurement data for the multiple cycles acquired by the data acquisition unit; and a sludge viscosity output unit that calculates and outputs a representative sludge viscosity of the activated sludge from the measurement data evaluated as normal by the evaluation unit.

[0021] The second characteristic configuration is, in addition to the first characteristic configuration described above, that the evaluation unit includes a machine learning device that has performed machine learning in advance based on teacher data that has been labeled with at least the measurement data when the amount of filtered water in the membrane separation device fluctuates, the measurement data when foreign matter is attached to the vibration viscometer, and the measurement data when the membrane separation device has a constant amount of filtered water and no foreign matter is attached to the vibration viscometer, and the evaluation unit evaluates the measurement data to be normal data if the label output when the measurement data acquired by the data acquisition unit is input to the machine learning device indicates normal times.

[0022] The third characteristic configuration is that, in addition to the first characteristic configuration described above, the evaluation unit evaluates the measurement data as normal data when it is determined that there are no outliers in the measurement data using LOF or One-Class SVM learned based on the measurement data of multiple cycles under normal conditions previously acquired by the data acquisition unit. [Effects of the Invention]

[0023] As described above, according to the present invention, it is possible to provide a viscosity measurement method and a viscosity measurement device that can properly detect the viscosity of sludge even if foreign matter is mixed into the sludge stored in the chamber. [Brief explanation of the drawings]

[0024] [Figure 1] An explanatory diagram of a biological treatment tank in which a membrane separation device is submerged and a viscosity measuring device for measuring the viscosity of activated sludge in the biological treatment tank. [Figure 2] Control device diagram [Figure 3] Flowchart showing the steps of the viscosity measurement method [Figure 4] (a) is an explanatory diagram of measurement data under normal (usual) conditions, (b) is an explanatory diagram of measurement data under abnormal conditions when foreign matter is entangled in the sensor, and (c) is an explanatory diagram of measurement data when the membrane filtration volume fluctuates. [Figure 5] An explanatory diagram of a neural network used as a machine learning device DETAILED DESCRIPTION OF THE INVENTION

[0025] The viscosity measuring method and viscosity measuring device according to the present invention will be described below. Figure 1 shows a biological treatment tank 1 that uses a membrane bioreactor (MBR) method, and a viscosity measuring device 9 that measures the viscosity of the activated sludge in the biological treatment tank 1. In the biological treatment tank 1, a membrane separation device 2 is installed immersed in the activated sludge, and an aeration device 5 is installed below the membrane separation device 2.

[0026] The membrane separation device 2 is configured such that multiple flat membrane modules are arranged in parallel in an up-down position inside a casing 3, and a flow path is formed between the membrane surfaces of adjacent flat membrane modules through which the upward flow generated by the aeration device 5 flows.

[0027] Each membrane module has a structure in which an external primary side and an internal secondary side are separated by a filtration membrane, and the secondary side is connected to a suction pump 4. A blower 6 is connected to an aeration device 5, and an upward flow is formed by air bubbles supplied from the air diffuser holes provided in the aeration device 5.

[0028] Organic wastewater such as sewage is supplied to the biological treatment tank 1 from a raw water supply system 7 connected to a raw water supply pump 8, and the organic components contained in the organic wastewater are decomposed and purified by activated sludge maintained in an aerobic state by air bubbles generated by the aeration device 5.

[0029] A sludge viscosity measuring device 9 is installed in the biological treatment tank 1. The sludge viscosity measuring device 9 includes an L-shaped chamber 10 that is filled with activated sludge from the biological treatment tank 1, a vibration type viscometer 11 that measures the viscosity of the activated sludge filled in the chamber 10, a sludge supply mechanism 12 that supplies activated sludge from the biological treatment tank 1 to the chamber 10, a cleaning liquid supply mechanism 13 that supplies cleaning liquid to the chamber 10, a sludge discharge mechanism 14 that discharges the activated sludge from the chamber 10 to the biological treatment tank 1, and a control device 15 that controls the sludge viscosity measuring device 9.

[0030] The chamber 10 is formed in an L-shape when viewed from the front, and a sludge supply mechanism 12 and a cleaning liquid supply mechanism 13 are connected to the lower end of the sludge supply path 10L in the lower area extending vertically, and a sludge discharge mechanism 14 is connected to the tip of the sludge discharge path 10U in the upper area extending horizontally.

[0031] The vibration viscometer 11 is installed in the curved path 10M of the chamber 10, that is, in the connecting area between the lower and upper areas. The vibration viscometer 11 is equipped with a probe 16 that is placed inside the chamber 10 and inserted into the sludge, and a measurement unit 17 that measures viscosity by vibrating the probe 16. The probe 16, which has an oscillator 16A at its tip, is covered with a cover that has an open tip and hangs downward in the lower area of ​​the chamber 10.

[0032] The cleaning liquid supply mechanism 13 is composed of a cleaning liquid supply pipe 13C connected to the sludge supply path 10L, a water supply pump P2 that supplies tap water as a cleaning liquid to the cleaning liquid supply pipe 13C, and a first valve V1 and a second valve V2 provided on the cleaning liquid supply pipe.

[0033] The sludge supply mechanism 12 is composed of a sludge supply pipe 12C that supplies sludge from the biological treatment tank 1 to the cleaning liquid supply mechanism 13, and a sludge supply pump P1 that supplies sludge to the sludge supply pipe 12C. The sludge discharge mechanism 14 is equipped with a sludge discharge pipe 14C that discharges sludge from the tip of the sludge discharge path 10U into the biological treatment tank 1.

[0034] As shown in Fig. 2, the control device 15 is configured with an electronic circuit including a control unit 15A, which is configured with an arithmetic circuit including a CPU, a memory 15B, an evaluation unit 15C, and input / output circuits. The control unit 15A controls the sludge supply pump P1, the water supply pump P2, the first valve V1, and the second valve V2 based on a control program stored in the memory 15B, and also controls the vibration viscometer 11 to measure viscosity. The evaluation unit 15C evaluates whether the measurement data input from the vibration viscometer 11 is normal or abnormal. If the evaluation unit 15C evaluates the measurement data as normal, the control unit 15A calculates and outputs a representative sludge viscosity of the activated sludge from the sludge viscosity data included in the measurement data.

[0035] FIG. 3 shows the viscosity measurement method carried out by the control device 15. The control unit 15A first closes the first valve V1 and opens the second valve V2, and then drives the sludge supply pump P1 to execute the sludge supply process A, in which activated sludge is supplied from the biological treatment tank 1 to the chamber 10 through the sludge supply pipe 12C (S1). The sludge supply step A is carried out, for example, once every hour, and in each execution, the sludge supply pump P1 is operated for one minute to fill the chamber 10 with activated sludge.

[0036] Next, with the chamber 10 filled with sludge, the second valve V2 is closed, the sludge supply pump P1 is stopped and held for one minute, and then a sludge viscosity measurement step B is executed (S2), in which the viscosity of the sludge filled in the chamber 10 is measured chronologically at a predetermined sampling period using the vibration viscometer 11. When the probe 16 is vibrated at a constant vibration frequency, viscous resistance acts between the probe 16 and the sludge, and the amplitude changes depending on the magnitude of the resistance. The measurement unit 17 measures the sludge viscosity based on the correlation between the sludge viscosity and the driving power required to vibrate the probe 16 at a constant vibration frequency.

[0037] Thereafter, with first valve V1 and second valve V2 open, water supply pump P2 is operated to inject tap water as a cleaning liquid for one minute, and the tap water is used to discharge sludge from inside chamber 10 via sludge discharge mechanism 14, while a cleaning step C is executed to clean the inside of chamber 10 and probe 16 (S3). At this time, probe 16 hanging down inside chamber 10 comes into contact with the tap water flow as a cleaning liquid in the axial direction, and the entire probe 16 is cleaned.

[0038] Furthermore, the first valve V1 and the second valve V2 are closed, the water supply pump P2 is stopped, tap water is stored inside the chamber 10, and the probe 16 is held in the tap water cleaning liquid. Then, a reference viscosity measurement step D is executed (S4), in which the vibration viscometer 11 measures the reference viscosity of the cleaning liquid in the chamber 10 in time series at a predetermined sampling period.

[0039] Next, the evaluation unit 15C evaluates the appropriateness of the measurement data based on the time-series measurement data consisting of a set of sludge viscosity and reference viscosity stored in memory 15B (S5). If the measurement data is evaluated as appropriate, i.e., normal data (S6, Y), a representative sludge viscosity is calculated from the time-series measurement data (S8), and the representative sludge viscosity is output as the measurement result (S9). If the evaluated time-series measurement data is determined to be abnormal (S6, N), a message indicating that the measurement data is abnormal is output. The processing from step S1 to step S9 is repeated at predetermined intervals.

[0040] FIG. 4(a) shows measurement data in a normal state, that is, a normal state. In the figure, the open circles plot the transmembrane pressure (TMP) of the membrane separation device 2 submerged in the biological treatment tank 1. The membrane separation device 2 repeats, at predetermined time intervals, a filtration operation in which treated water is drawn from the membrane separation device 2 using the suction pump 4 while aeration is performed using the aeration device 5, and a relaxation operation in which the suction pump 4 is stopped while aeration by the aeration device 5 is maintained, thereby purifying the membrane. For example, a 9-minute filtration operation and a 1-minute relaxation operation are repeated. The transmembrane pressure (TMP), which rises during the filtration operation, falls during the relaxation operation. Note that the times for the filtration operation and relaxation operation are not limited to these values ​​and can be set as appropriate.

[0041] In the figure, the black circles are plots of time-series viscosity measurement data measured by the vibration viscometer 11. The black circles that indicate a viscosity of approximately 1 indicate the reference viscosity, which is the viscosity of the cleaning liquid, and the black circles that increase significantly relative to the reference viscosity and then saturate indicate the sludge viscosity. A sludge viscosity measurement, in which the sludge viscosity is measured over time for the activated sludge filled in the chamber 10, and a reference viscosity measurement, in which the reference viscosity is measured over time for the cleaning liquid filled in the chamber 10, are repeated alternately as one cycle, thereby obtaining multiple cycles of measurement data, each consisting of a set of sludge viscosity and reference viscosity obtained over time. For example, one cycle from the measurement of the sludge viscosity to the measurement of the reference viscosity is repeated every hour. The duration of one cycle is not limited to one hour and can be set as appropriate.

[0042] FIG. 4(b) shows measurement data in an abnormal state where a foreign substance is attached to the probe 16. In the figure, the open circles represent the transmembrane pressure difference TMP of the membrane separation device 2 immersed in the biological treatment tank 1, and the closed circles represent the time-series viscosity measurement data measured by the vibration viscometer 11.

[0043] The black dots indicating a viscosity of approximately 1 are the reference viscosity measured in the reference viscosity measurement process, and the black dots showing a monotonically increasing trend are the sludge viscosity measured in the sludge viscosity measurement process. The sludge viscosity measurement value monotonically increases without converging due to the influence of foreign matter entangled on the probe in the sludge viscosity measurement process. Note that if foreign matter is not removed from the probe 16 in the cleaning process, the reference viscosity will not decrease either.

[0044] Figure 4(c) shows the behavior of the base viscosity and sludge viscosity when TMP rises and membrane filtration is stopped for an extended period of time in addition to a one-minute relaxation operation during membrane purification. In this example, one cycle from sludge viscosity measurement to base viscosity measurement is repeated every 0.5 hours. In the figure, the dashed line indicates the fluctuations in MLSS, which can be seen to decrease around 6:00 and 19:00. The MLSS decreased because the withdrawal of treated water from membrane separation device 2 was stopped for about an hour during these periods. As a result, a tendency for the saturated value of sludge viscosity to transiently decrease compared to normal membrane separation operation is observed.

[0045] Similarly, when the amount of treated water withdrawn from the membrane separation device 2 is increased or decreased, the saturated value of sludge viscosity tends to transiently decrease or increase relative to the normal amount withdrawn. If the clogging state of the membrane separation device 2 provided in the membrane separation device 2 is judged based on the sludge viscosity measured under the abnormal conditions described above, it may lead to an erroneous judgment of the clogging state of the membrane, which may result in over-aeration or under-aeration by the aeration device 5, or an erroneous judgment of the timing of membrane purification.

[0046] The evaluation unit 15C is provided to prevent such erroneous determination and to properly grasp the sludge viscosity. The evaluation unit 15C automatically evaluates whether the data is normal or abnormal based on the fluctuation characteristics of the measurement data acquired over multiple cycles in the data acquisition process. That is, the evaluation unit 15C compares multiple cycles of measurement data, each set consisting of a sludge viscosity and a reference viscosity acquired over time, to determine whether the sludge viscosity is fluctuating within a range showing appropriate saturation characteristics and whether the reference viscosity has decreased to an appropriate viscosity relative to the sludge viscosity.

[0047] For this reason, the evaluation unit 15C incorporates a machine learning device using a neural network, or a machine learning device using LOF or One-Class SVM.

[0048] Figure 5 shows a machine learning device using a neural network. The neural network is composed of three layers: an input layer with n nodes, a hidden layer with m nodes, and an output layer with three nodes. A weight vector w1 is set between the input layer and hidden layer, and a weight vector w2 is set between the hidden layer and the output layer. Here, n and m are natural numbers. When a feature vector x (x1, x2, . . . , xn) is input as input data to a node in the input layer, the operation shown in the following equation is executed at each node, and the result is input to the next node, and finally, a feature vector z (z1, z2, z3) is output as output data from a node in the output layer.

[0049] Each hidden layer node receives an input xi from each node in the input layer, calculates one output y based on the calculation shown in the following equation, and outputs it to each node in the subsequent stage. y=φ(Σxi·w1i+θ) The nodes in the output layer receive input yj from each node in the intermediate layer, and calculate and output one output z based on the calculation shown in the following equation. z=φ(Σyi·w2j+θ) Here, θ is the bias and φ is the activation function, which can be a step function, a simple perceptron, a sigmoid function, or the like.

[0050] The neural network learns weight vectors w1 and w2 based on pre-prepared input training data and output training data so that the difference between the output from the output layer and the output training data is minimized. Backpropagation is a suitable learning algorithm for the weight vectors w1 and w2. Other methods for adjusting the weight vectors w1 and w2 include stochastic gradient descent, RMSprop, and Adamax.

[0051] In this embodiment, n cycles of measurement data (teacher data 1) showing the behavior of the membrane separation device when the amount of filtered water fluctuates, n cycles of measurement data (teacher data 2) showing the behavior when foreign matter adheres to the vibration viscometer, and n cycles of measurement data (teacher data 3) showing the behavior of the membrane separation device under normal conditions when the amount of filtered water is constant and no foreign matter adheres to the vibration viscometer are used as input teacher data, and teacher data is prepared in which the output teacher data is a combination of the corresponding labels 1 or 0. For teacher data 1, the output teacher data (z1, z2, z3) is (1, 0, 0), for teacher data 2 it is (0, 1, 0), and for teacher data 3 it is (0, 0, 1).

[0052] That is, when training data for each cycle is input to n nodes in the input layer, the weight vectors w1 and w2 are adjusted so that corresponding labels are output from three nodes in the output layer.

[0053] When actual n-cycle measurement data is input to the input layer of the neural network trained in this way, the corresponding label is output from the output layer. For example, if the output vector (z1, z2, z3) is close to (0, 0, 1), it is considered normal, and if it is close to (0, 1, 0), it is considered that foreign matter has adhered to the vibration viscometer.

[0054] When the evaluation unit 15C determines that the measurement data for n cycles is normal, the control unit 15A extracts the saturated values ​​from the sludge viscosity included in each measurement data for the n cycles, calculates the average value thereof as the representative sludge viscosity, and outputs it. Then, when the evaluation unit 15C determines that the measurement data for the nth cycle is abnormal, the control unit 15A outputs a signal indicating an abnormal state corresponding to the value of the output vector at that time without calculating the representative sludge viscosity.

[0055] The value of the number of nodes n in the input stage is not particularly limited as long as it matches the number of cycles of the measurement data, and may be set appropriately based on the time of one cycle of the data acquisition process.

[0056] An example will be described in which the evaluation unit 15C is equipped with a machine learning device using LOF. LOF is one of the outlier detection algorithms that finds outliers from a predetermined collection of data, and is a method for determining whether a point of interest is included in the same data group using the local density of k neighboring points from the point of interest in space as an index. The k neighbors refer to the k points that are closest to the point of interest, and the local density refers to the reciprocal of the average value of the distances to the k neighboring points. The more the local density of the point of interest and the local density of the neighboring points are equal, the more normal the data can be evaluated, and the greater the difference, the more it can be evaluated as an outlier.

[0057] For example, for each of the sludge viscosity and reference viscosity included in one cycle of measurement data, the local density for k neighboring points for each measurement value is calculated, and if the difference between the local density for each target point and the local density of the neighboring point is within a preset threshold, the measurement data is evaluated as normal. This process is repeated multiple times, and if all results are determined to be normal, the measurement data can be evaluated as normal data without any abnormalities. In an embodiment such as FIG. 4(b), evaluation can be performed based on one cycle of measurement data, but in an embodiment such as FIG. 4(b), evaluation must be performed based on multiple cycles of measurement data. The value of k is set appropriately depending on the characteristics of the measurement data, and in this embodiment, the value of k is set to 4.

[0058] An example will be described in which the evaluation unit 15C is provided with a machine learning device using a one-class support vector machine (One-Class SVM). The one-class support vector machine is a method that applies support vector machines (SVMs) to unsupervised one-class classification, and is an algorithm that learns one class of normal data, determines a classification boundary, and then detects outliers based on that boundary.It is an outlier detection method that can effectively detect anomalies when it is difficult to collect training data that indicates abnormal conditions.

[0059] All measurement data, or training data, is placed in cluster 1, and a method called the kernel method is used to map the data into a high-dimensional feature space so that only the origin belongs to cluster 1. Because the training data is mapped so that it is placed far from the origin, data that is not similar to the original training data will gather near the origin. This property is used to evaluate whether the measurement data is in a normal state, that is, data in a normal state, or abnormal data.

[0060] That is, the viscosity measuring device that performs the above-mentioned viscosity measurement method includes a data acquisition unit that acquires multiple cycles of measurement data, each consisting of a pair of sludge viscosity and reference viscosity, obtained in a time series by alternately repeating one cycle of sludge viscosity measurement, which measures the sludge viscosity of activated sludge filled in a chamber equipped with a vibration viscometer, and a reference viscosity measurement, which measures the reference viscosity of cleaning liquid filled in the chamber in a time series; an evaluation unit that evaluates whether the data is normal or abnormal based on the fluctuation characteristics of the measurement data obtained in the multiple cycles by the data acquisition unit; and a sludge viscosity output unit that calculates and outputs a representative sludge viscosity of the activated sludge from the measurement data evaluated as normal by the evaluation unit.

[0061] The evaluation unit is equipped with a machine learning device that has previously performed machine learning based on teacher data that has been labeled with at least measurement data when the amount of filtered water in the membrane separation device fluctuates, measurement data when foreign matter adheres to the vibration viscometer, and measurement data under normal conditions when the membrane separation device has a constant amount of filtered water and no foreign matter adheres to the vibration viscometer, and the evaluation unit evaluates the measurement data as normal data if the label output when the measurement data acquired by the data acquisition unit is input into the machine learning device indicates normal conditions.

[0062] The evaluation unit is a device that uses LOF or One-Class SVM learned based on the measurement data of multiple cycles under normal conditions previously acquired by the data acquisition unit, and evaluates the measurement data as normal data when it is determined that there are no outliers in the measurement data.

[0063] The above-described embodiment is one aspect of the present invention, and the present invention is not limited to this description. It goes without saying that the specific configuration of each part can be appropriately modified and designed within the scope of the effects of the present invention. [Explanation of symbols]

[0064] 1: Biological treatment tank 2: Membrane separation device 3: Fine mesh screen 4: Suction pump 5: Aeration device 9: Sludge viscosity measuring device 12: Sludge supply mechanism 13: Cleaning liquid supply mechanism 14: Sludge discharge mechanism 15: Control device 15A: Control unit 15C: Evaluation unit (machine learning device) 16: Probe

Claims

1. A viscosity measurement method for measuring the viscosity of activated sludge filled in a biological treatment tank in which a membrane separation device is immersed, comprising: a data acquisition process in which a sludge viscosity measurement is performed in a time-series manner on the activated sludge filled in a chamber equipped with a vibration viscometer, and a reference viscosity measurement is performed in a time-series manner on the cleaning liquid filled in the chamber, and this is repeated alternately as one cycle to acquire measurement data of multiple cycles, each set consisting of the sludge viscosity and the reference viscosity acquired in a time-series manner; an evaluation step of evaluating whether the data is normal or abnormal based on fluctuation characteristics of the measurement data for multiple cycles acquired in the data acquisition step; a sludge viscosity output step of calculating and outputting a representative sludge viscosity of the activated sludge from the measurement data evaluated as normal data in the evaluation step; A viscosity measurement method comprising:

2. The evaluation step includes a machine learning device that has previously performed machine learning based on teacher data that is labeled with at least the measurement data when the amount of filtrate in the membrane separation device fluctuates, the measurement data when foreign matter adheres to the vibration viscometer, and the measurement data when the membrane separation device has a constant amount of filtrate and no foreign matter adheres to the vibration viscometer, The viscosity measurement method according to claim 1, wherein the evaluation step is a step of evaluating the measurement data as normal data if the label output when the measurement data acquired in the data acquisition step is input into the machine learning device indicates normal times.

3. 2. The viscosity measurement method according to claim 1, wherein the evaluation step is a step of evaluating the measurement data as normal data when it is determined that no outliers exist in the measurement data using LOF or One-Class SVM learned based on the measurement data of multiple cycles under normal conditions acquired before the data acquisition step.

4. 4. The viscosity measuring method according to claim 1, wherein the representative viscosity output in the sludge viscosity output step is an average value of saturated values ​​of the sludge viscosity included in the measurement data for the plurality of cycles.

5. A viscosity measuring device for measuring the viscosity of activated sludge filled in a biological treatment tank in which a membrane separation device is immersed, a data acquisition unit that acquires measurement data of multiple cycles, each set consisting of the sludge viscosity and the reference viscosity, acquired in a time series manner by alternately repeating one cycle of sludge viscosity measurement, in which the activated sludge filled in a chamber equipped with a vibration viscometer is measured in a time series manner, and a reference viscosity measurement, in which the cleaning liquid filled in the chamber is measured in a time series manner; an evaluation unit that evaluates whether the data is normal or abnormal based on fluctuation characteristics of the measurement data over multiple cycles acquired by the data acquisition unit; a sludge viscosity output unit that calculates and outputs a representative sludge viscosity of the activated sludge from the measurement data that has been evaluated as normal data by the evaluation unit; A viscosity measuring device comprising:

6. The evaluation unit includes a machine learning device that has previously performed machine learning based on teacher data that has been labeled with at least the measurement data when the amount of filtrate in the membrane separation device fluctuates, the measurement data when foreign matter adheres to the vibration viscometer, and the measurement data when the membrane separation device has a constant amount of filtrate and no foreign matter adheres to the vibration viscometer, The viscosity measurement device according to claim 5, wherein the evaluation unit evaluates the measurement data as normal data if the label output when the measurement data acquired by the data acquisition unit is input into the machine learning device indicates normal times.

7. 2. The viscosity measurement method according to claim 1, wherein the evaluation unit evaluates the measurement data as normal data when it is determined that no outliers exist in the measurement data using LOF or One-Class SVM learned based on the measurement data of multiple cycles under normal conditions previously acquired by the data acquisition unit.

Images