JP5896105B2 - pH measuring device - Google Patents

Description

  The present invention relates to a reference pH calibration solution that is executed at predetermined time intervals or at any time with respect to pH conversion calculation means for inputting a measurement value of a glass electrode type pH sensor whose pH measurement sensitivity of a solution to be measured changes with time. The present invention relates to a pH measurement apparatus that passes calibration sensitivity data acquired in the calibration work by the method and corrects the pH measurement sensitivity.

  FIG. 6 is a functional block diagram showing a configuration example of a conventional pH measurement device. In the measuring apparatus 10, the measured value X of the measuring liquid 12 by the pH sensor 11 is input to the pH conversion calculating means 13 and is calculated and converted into a pH value.

  When the measurement sensitivity of the pH sensor is a and the bias value indicating the zero point is b, the output pH of the pH conversion calculation means 13 is pH = aX + b. However, the measurement sensitivity a of the pH sensor changes with the operating time.

  In order to compensate for this change in sensitivity, the correction value is calculated by the sensitivity correction means 13a based on the calibration sensitivity a ′ (t) acquired by calibration of the pH sensor 11 performed at a predetermined time interval or at any time, pH = a ′ ( t) A process for outputting X + b is required. Calibration is also required when the bias value b fluctuates with time, but for the sake of simplicity of explanation, it will be explained here as a fixed constant.

  The calibration device 20 includes a calibration liquid 22 having a known pH value. A pH conversion calculation output pH = a ′ (t obtained by inputting the measured value X obtained by discretely measuring the calibration liquid 22 by the pH sensor 11 to the pH conversion calculation means 23 at a predetermined time interval or at any time. ) X + b is passed to the calibration sensitivity extraction means 24 to extract the calibration sensitivity a ′ (t).

  The past calibration sensitivity a ′ (t) obtained discretely extracted by the calibration sensitivity extraction unit 24 is stored in the calibration history storage unit 25. The sensitivity change curve prediction unit 26 reads the trend data stored in the calibration history storage unit 25, and based on the calibration date and the sensitivity history data at that time, the sensitivity that is an average time course versus sensitivity change approximate curve. Guide the change curve. The purpose of this sensitivity change curve is to derive a curve closer to the actual sensitivity change.

  Therefore, this curve may be a straight line or a multi-order curve having many orders. In deriving an approximate curve, the method is such that linear approximation is performed using the two most recent calibration data, or linear approximation is performed using the least square approximation method using the latest three or more calibration data. Several methods are conceivable.

  The pH calculation conversion means 13 of the measuring apparatus 10 inputs the sensitivity correction value a ″ (t) obtained continuously from the sensitivity change curve predicted by the sensitivity change curve prediction means 26, and the sensitivity correction is performed by the sensitivity correction means 13a. , PH = a ″ (t) X + b is calculated and output.

  FIG. 7 is a characteristic diagram showing a sensitivity change curve of the conventional pH measuring apparatus shown in FIG. An actual sensitivity change curve is indicated by a dotted line F1 from the calibration sensitivity a1 at the operation start time t1. The calibration sensitivities acquired by the calibration performed at times t2 and t3 are denoted by a2 and a3.

  The sensitivity curve obtained by calibration is indicated by a solid line F2. F2 maintains the calibration sensitivity a1 acquired at t1 from time t1 to t2, maintains the calibration sensitivity a2 acquired at t2 from time t2 to t3, and so on. Data.

  The sensitivity change curve predicted by the sensitivity change curve prediction unit 26 is F3 indicated by a bold solid line. The sensitivity change curve F3 shown in the figure is approximated by a straight line connecting the calibration sensitivity a2 obtained at time t2 and the calibration sensitivity a3 obtained at time t3.

  Since the predicted sensitivity change curve F3 and the actual sensitivity change curve F1 are close to each other, the sensitivity deviation Δa ′ generated at time t4 before the next calibration timing is made smaller than the maximum deviation Δa indicated by the discrete data. be able to.

Japanese Patent Laid-Open No. 5-164736

  In the conventional pH measuring apparatus, as an interpolation function for discrete calibration data, calibration data is predicted from the accumulated calibration data using an approximate expression. In this method, when a glass electrode type pH sensor is used as the pH sensor, the impedance change and the temperature change of the measurement solution, which are degradation factors, are not considered. There is a limit to reducing the deviation. In other words, since the approximation is based on accumulated data, it is difficult to eliminate the calibration work itself.

  An object of the present invention is to realize a pH measuring device that can reduce the deviation of the conventional forward prediction function of the sensitivity change, greatly reduce the number of calibrations, or eliminate calibration.

In order to achieve such a subject, the present invention has the following configuration.
(1) For a pH conversion calculation means for inputting a measured value of a glass electrode pH sensor in which the pH measurement sensitivity of the liquid to be measured changes with time, using a calibration solution for a reference pH that is executed at predetermined time intervals or as needed. In the pH measurement device for correcting the pH measurement sensitivity by passing calibration sensitivity data acquired in the calibration operation,
The elapsed time, the pH measurement value of the pH conversion calculation means, the temperature measurement value of the measurement liquid, and the impedance measurement value of the glass electrode type pH sensor are used as learning data, and back of calibration sensitivity data acquired in the calibration operation. Sensitivity coefficient reasoning means for inputting data as teacher data;
Impedance detection means for measuring the impedance of the glass electrode type pH sensor during pH measurement of the liquid to be measured and providing the measured value to the sensitivity coefficient inference means online ;
With
The sensitivity coefficient inference means predicts the calibration sensitivity between the acquisition of the calibration sensitivity data and the next acquisition of the calibration sensitivity data, and passes it as calibration sensitivity data to the pH conversion calculation means. apparatus.

(2) In the offline state, the sensitivity coefficient inference means uses the elapsed time, the pH measurement value of the pH conversion calculation means, the temperature measurement value of the measurement liquid, and the impedance measurement value of the glass electrode pH sensor as learning data. The pH measurement apparatus according to (1), wherein back data of calibration sensitivity data acquired in the calibration operation is input as teacher data to learn an inference algorithm.

(3) In the online state, the sensitivity coefficient inference means uses the elapsed time, the pH measurement value of the pH conversion calculation means, the temperature measurement value of the measurement liquid, and the impedance measurement value of the glass electrode pH sensor as learning data. The pH measurement apparatus according to (1) or (2), wherein back data of calibration sensitivity data acquired in the calibration operation is input as teacher data to learn an inference algorithm.

(4) The pH measurement device according to any one of (1) to (3), wherein the sensitivity coefficient inference means is configured by a neural network.

According to the present invention, the following effects can be expected.
(1) Predicting the change in sensitivity by means of sensitivity coefficient reasoning means that can be realized by a neural network or the like using the degradation factor of the pH sensor as learning data and the calibration result data as teacher data, The deviation can be made smaller.

(2) If the accuracy of the back data can be ensured, sensitivity can be predicted without calibration after performing calibration several times to learn the installed environment and application.

It is a functional block diagram which shows one Example of the pH measuring apparatus to which this invention is applied. It is a schematic diagram explaining the learning operation | movement by a neural network. It is a schematic diagram explaining the prediction operation in operation by the neural network. It is an impedance measurement circuit diagram of a glass electrode pH sensor. It is a characteristic view which shows the sensitivity change curve of the pH measuring apparatus by this invention. It is a functional block diagram which shows the structural example of the conventional pH measuring apparatus. It is a characteristic view which shows the sensitivity change curve of the conventional pH measuring apparatus.

  Hereinafter, the present invention will be described in detail with reference to the drawings. FIG. 1 is a functional block diagram showing an embodiment of a pH measuring device to which the present invention is applied. The same elements as those in the conventional configuration described with reference to FIG.

  The feature of the present invention added to the measurement apparatus 10 having the conventional configuration shown in FIG. 6 is sensitivity coefficient inference for forward prediction of the sensitivity coefficient a ″ (t) passed to the sensitivity correction means 13 a of the pH conversion calculation means 13. The means 100 is provided.

  A known neural network can be adopted as the sensitivity coefficient inference means 100, but is not limited to this, and has been put into practical use for predicting process properties that are difficult to measure directly from a large number of process values. It is possible to employ a known multivariable model predictive control package. An embodiment using a neural network will be described below.

  Sensitivity coefficient inference means 100 using a neural network learns deterioration factors such as elapsed time t, output pH value of pH conversion calculation means, temperature measurement value T of measurement liquid 12, and glass electrode impedance measurement value Z of pH sensor 11. While inputting as data, back data a ′ (t) of sensitivity data obtained from the calibration history storage unit 25 is input as teacher data, and a predicted sensitivity coefficient a ″ (t) is output, to the sensitivity correction unit 13a. hand over.

  The glass electrode impedance measurement value Z is measured by the periodic detection process by the impedance detection means 14. What is the temperature measurement T? The temperature is measured by temperature detection means 16 using a temperature sensor 15 in contact with the measurement liquid 12.

  FIG. 2 is a schematic diagram for explaining a learning operation by the neural network. Sensitivity coefficient inference means 100 using a neural network inputs elapsed time t, output pH value of pH conversion calculation means, temperature measurement value T of measurement liquid 12, and glass electrode impedance measurement value Z of pH sensor 11 as learning data parameters. At the same time, the back data a ′ (t) of the sensitivity data is input as teacher data, and learning is performed to optimize the inference algorithm.

  As a learning method of the neural network, the back data of the sensitivity change with respect to the elapsed time, pH measurement value, temperature measurement value, impedance measurement value, etc. which are input parameters in the offline state during development is measured, and the back data of the sensitivity change is instructed. Learn as data.

  This back data is preferably learned with as much data as possible. In addition, every time the user calibrates while operating online, the user learns from the proofreading data so that it can cope with an environment specialized for the application.

  FIG. 3 is a schematic diagram for explaining a prediction operation during operation by the neural network. As input data to the sensitivity coefficient inference means 100 by the neural network, the same elapsed time t as the offline learning data, the output pH value of the pH conversion calculation means, the temperature measurement value T of the measurement liquid 12, the glass electrode of the pH sensor 11 The impedance measurement value Z is input online, the back data a ′ (t) of sensitivity data is input as online teacher data, the predicted sensitivity coefficient a ″ (t) is calculated online as output data, and sensitivity correction means Pass to 13a.

  FIG. 4 is an impedance measurement circuit diagram of the glass electrode pH sensor. An equivalent circuit of the glass electrode pH sensor can be expressed by a series circuit of an electromotive force Vg and an impedance Z connected between the liquid earth and the glass electrode. A 100Hz square wave signal V is applied to the liquid earth side in response to a command from the CPU so as not to affect the DC electromotive force in the pH measurement. On the electrode side, the impedance Z to be measured and the partial voltage Vin of the fixed resistor R are sampled and held. The converted voltage Vo is AD converted and input to the CPU.

  FIG. 5 is a characteristic diagram showing a sensitivity change curve of the pH measuring device according to the present invention. As is clear from the contrast between the sensitivity change curve F3 according to the prediction of the conventional configuration shown in FIG. 7 and the sensitivity change curve F3 ′ according to the prediction of the present invention, the sensitivity deviation Δa ′ at time t4 becomes smaller and the accuracy of the prediction is reduced. Has improved.

  According to the method of the present invention, the predicted sensitivity change curve F3 ′ can be approximated with high accuracy to the actual sensitivity change curve F1. Thereby, the frequency of calibration can be reduced as compared with the conventional configuration, and it is also possible to shift to a calibration-less operation after performing learning for a predetermined period.

DESCRIPTION OF SYMBOLS 10 Measuring apparatus 11 pH sensor 12 Measuring liquid 13 pH conversion calculating means 13a Sensitivity correction means 14 Impedance detecting means 15 Temperature sensor 16 Temperature detecting means 20 Calibration apparatus 22 Calibration solution 23 pH conversion calculating means 24 Calibration sensitivity extracting means 25 Calibration history saving means 25 100 Sensitivity coefficient reasoning means

Claims (4)

For the pH conversion calculation means for inputting the measured value of the glass electrode type pH sensor whose pH measurement sensitivity of the liquid to be measured changes with time, the calibration work with the calibration solution of the reference pH that is executed at a predetermined time interval or at any time In a pH measurement device that passes the acquired calibration sensitivity data and corrects the pH measurement sensitivity,
The elapsed time, the pH measurement value of the pH conversion calculation means, the temperature measurement value of the measurement liquid, and the impedance measurement value of the glass electrode type pH sensor are used as learning data, and back of calibration sensitivity data acquired in the calibration operation. Sensitivity coefficient reasoning means for inputting data as teacher data;
Impedance detection means for measuring the impedance of the glass electrode type pH sensor during pH measurement of the liquid to be measured and providing the measured value to the sensitivity coefficient inference means online ;
With
The sensitivity coefficient inference means predicts the calibration sensitivity between the acquisition of the calibration sensitivity data and the next acquisition of the calibration sensitivity data, and passes it as calibration sensitivity data to the pH conversion calculation means. apparatus.   The sensitivity coefficient inference means uses the elapsed time in the offline state, the pH measurement value of the pH conversion calculation means, the temperature measurement value of the measurement liquid, and the impedance measurement value of the glass electrode pH sensor as learning data, and 2. The pH measuring apparatus according to claim 1, wherein back data of calibration sensitivity data acquired in a calibration operation is input as teacher data to learn an inference algorithm.   The sensitivity coefficient inference means uses the elapsed time in the online state, the pH measurement value of the pH conversion calculation means, the temperature measurement value of the measurement liquid, and the impedance measurement value of the glass electrode type pH sensor as learning data, and The pH measuring apparatus according to claim 1 or 2, wherein the back data of the calibration sensitivity data acquired in the calibration operation is input as teacher data to learn an inference algorithm.   4. The pH measuring device according to claim 1, wherein the sensitivity coefficient inference means is constituted by a neural network.