JP2025174068A - Calibration curve generating device, calibration curve generating method, and calibration curve generating program - Google Patents

Abstract

A calibration curve generating device, a calibration curve generating method, and a calibration curve generating program are provided that can prevent an inspection device from erroneously determining inspection results.
[Solution] A calibration curve generating device 2 generates a calibration curve to be referenced by an inspection device 11 that inspects an article based on the spectroscopic spectrum of the article measured by a spectroscopic measurement device 10, and includes a preprocessing unit 20 that preprocesses multiple absorption spectra obtained by the spectroscopic measurement device 10 for multiple articles whose true values are known, a classification unit 21 that classifies the multiple preprocessed absorption spectra into teacher data and test data, and a calibration curve calculation unit 22 that calculates a calibration curve from the teacher data and the true values. The classification unit 21 calculates statistical values obtained from each preprocessed absorption spectrum, and selects absorption spectra to be classified into teacher data so as to evenly cover all the statistical values.
[Selected Figure] Figure 1

Description

The present invention relates to a calibration curve generating device, a calibration curve generating method, and a calibration curve generating program.

Patent Document 1 discloses an inspection device that detects whether the components of a molded product are appropriate by directing the transmitted light emitted from a light source that has passed through the molded product into a sensor as signal light and analyzing that signal light.

Japanese Patent Application Laid-Open No. 2019-112199

One type of inspection device, such as that described in Patent Document 1, inspects whether the components of a molded product are appropriate based on a calibration curve. The calibration curve is generated from training data whose true values are known. However, if the calibration curve is generated from training data with low comprehensiveness, the accuracy of the calibration curve will decrease, which may cause the inspection device to erroneously determine the inspection results.

The present invention was made in consideration of the above-mentioned circumstances, and aims to provide a calibration curve generation device, calibration curve generation method, and calibration curve generation program that can prevent the testing device from erroneously determining the test results.

The calibration curve generation device of the present invention is a calibration curve generation device (2) that generates a calibration curve referenced by an inspection device (11) that inspects an item based on the spectroscopic spectrum of the item measured by a spectroscopic measurement device (10). It includes a preprocessing unit (20) that preprocesses multiple absorption spectra obtained by the spectroscopic measurement device for multiple items whose true values are known, a classification unit (21) that classifies the preprocessed multiple absorption spectra into teacher data and test data, a calibration curve calculation unit (22) that calculates the calibration curve from the teacher data and the true values, and a calibration curve evaluation unit (23) that evaluates the calibration curve from the test data and the true values. The classification unit is configured to calculate statistical values obtained from each preprocessed absorption spectrum and select absorption spectra to be classified as teacher data so as to evenly cover all of the statistical values.

With this configuration, the calibration curve generation device of the present invention improves the accuracy of the calibration curve by selecting absorption spectra to be classified as training data so as to provide high comprehensiveness for the statistical values obtained from each absorption spectrum, thereby preventing the testing device from making erroneous judgments about the test results.

In the calibration curve generating device according to the present invention, the statistical values may include a first type of statistical values and a second type of statistical values, and the classifier may select absorption spectra to be classified into the training data so as to evenly cover the first type of statistical values, and may select absorption spectra to be classified into the training data from the absorption spectra excluding the absorption spectra classified into the training data from the plurality of absorption spectra so as to evenly cover the second type of statistical values.

With this configuration, the calibration curve generation device of the present invention improves the accuracy of the calibration curve by selecting absorption spectra to be classified as training data so as to achieve high comprehensiveness for the multiple statistical values obtained from each absorption spectrum, thereby preventing the testing device from making erroneous judgments about the test results.

Furthermore, in the calibration curve generating device according to the present invention, the statistical value may be any of the average value, maximum value, minimum value, standard deviation, and range.

With this configuration, the calibration curve generation device of the present invention improves the accuracy of the calibration curve by selecting absorption spectra to be classified into training data so that the statistical values obtained from each absorption spectrum are highly comprehensive with respect to the mean value, maximum value, minimum value, standard deviation, or range, thereby preventing the testing device from making erroneous judgments about the test results.

The calibration curve generation method of the present invention generates a calibration curve to be referenced by an inspection device (11) that inspects an item based on the spectroscopic spectrum of the item measured by a spectroscopic measurement device (10). The calibration curve generation method includes a preprocessing step of preprocessing multiple absorption spectra obtained by the spectroscopic measurement device for multiple items whose true values are known, a classification step of classifying the preprocessed absorption spectra into teacher data and test data, a calibration curve calculation step of calculating the calibration curve from the teacher data and the true values, and a calibration curve evaluation step of evaluating the calibration curve from the test data and the true values. The classification step calculates statistical values from each preprocessed absorption spectrum and selects absorption spectra to be classified as teacher data so as to evenly cover all of the statistical values.

In this way, the calibration curve generation method according to the present invention improves the accuracy of the calibration curve by selecting absorption spectra to be classified as training data so as to provide high comprehensiveness for the statistical values obtained from each absorption spectrum, thereby preventing the testing device from making erroneous judgments about the test results.

The calibration curve generation program of the present invention causes a computer to generate a calibration curve to be referenced by an inspection device (11) that inspects an item based on the spectroscopic spectrum of the item measured by a spectroscopic measurement device (10). The calibration curve generation program includes a preprocessing step that preprocesses multiple absorption spectra obtained by the spectroscopic measurement device for multiple items whose true values are known; a classification step that classifies the preprocessed absorption spectra into teacher data and test data; a calibration curve calculation step that calculates the calibration curve from the teacher data and the true values; and a calibration curve evaluation step that evaluates the calibration curve from the test data and the true values. The classification step calculates statistical values from each preprocessed absorption spectrum and selects absorption spectra to be classified as teacher data so as to evenly cover all of the statistical values.

In this way, the calibration curve generation program of the present invention improves the accuracy of the calibration curve by selecting absorption spectra to be classified as training data so as to provide high comprehensiveness for the statistical values obtained from each absorption spectrum, thereby preventing the testing device from making erroneous judgments about the test results.

The present invention provides a calibration curve generation device, a calibration curve generation method, and a calibration curve generation program that can prevent the testing device from erroneously determining the test results.

FIG. 1 is a functional block diagram of an article inspection apparatus equipped with a calibration curve generating device according to an embodiment of the present invention. FIG. 2 is a conceptual diagram showing the number of absorption spectra classified into training data by a classification unit included in a calibration curve generating device according to one embodiment of the present invention. FIG. 3 is a functional block diagram showing in detail the input and output of a calibration curve calculation unit that constitutes a calibration curve generating device according to one embodiment of the present invention. FIG. 4 is a functional block diagram showing in detail the input and output of a calibration curve evaluation unit that constitutes a calibration curve generating device according to one embodiment of the present invention. FIG. 5 is a conceptual diagram showing an example of a display for confirming the manner of classification by the classification unit constituting the calibration curve generating device according to one embodiment of the present invention.

Below, with reference to the drawings, we will explain an article inspection system equipped with a calibration curve generation device according to one embodiment of the present invention. As shown in Figure 1, the article inspection system according to this embodiment includes an article inspection device 1 and a calibration curve generation device 2.

When an item to be inspected, which is being transported individually along a transport path by a transport unit, reaches a predetermined inspection position, the item inspection device 1 irradiates light onto the item, which is in a fixed position at the predetermined inspection position, and inspects the quality of the item based on the spectral spectrum of the light (also called irradiated light) that passes through the item upon irradiation.

Items subject to inspection include bite-sized items with an outer diameter of several to several tens of millimeters that can be transported individually without packaging, as well as items and molded products of a predetermined shape manufactured using existing manufacturing equipment or manufacturing equipment without inspection capabilities, and especially items whose shape does not change during transportation.

Applicable articles include, for example, tablets, capsules, lozenges, drops, and other pharmaceutical preparations, as well as candy and chocolate. Below, we will use as an example a tablet W that is circular in plan view, has a height (thickness) smaller than its diameter, and is roughly cylindrical in side view. Note that the articles to be inspected are not limited to those that are circular in plan view, and can be articles of various shapes, such as ovals and polygons.

Examples of conveying units include conveying units configured to align and convey items individually, such as conveying belts, conveying disks, and conveying chutes.

The item inspection device 1 according to this embodiment is configured to include a spectroscopic measurement device 10 and an inspection device 11.

The spectroscopic measurement device 10 includes a light source unit and a light detection unit. The spectroscopic measurement device 10 irradiates a tablet W to be measured with broadband light (visible light, near-infrared to terahertz light (terahertz waves)) and measures the optical spectrum of the light that passes through the tablet W as it is irradiated with this light.

In this embodiment, the inspection device 11 is composed of a computer unit equipped with a CPU (Central Processing Unit), RAM (Random Access Memory), ROM (Read Only Memory), a storage device such as a hard disk drive, and a communication port.

The ROM and storage device of this computer unit store programs that cause the computer device to function as the inspection device 11. In other words, the CPU uses the RAM as a working area to execute the programs stored in the ROM and storage device, causing the computer unit to function as the inspection device 11 in this embodiment.

The inspection device 11 calculates the absorption spectrum of the tablet W by calculating the absorbance A of the tablet W at each wavelength λ according to A = -log ₁₀ (I/Ii) from the light intensity Ii at wavelength λ of the reference spectrum stored in the storage device and the light intensity I at wavelength λ of the spectrum of the tablet W measured by the spectroscopic measurement device 10. In this embodiment, the spectroscopic spectrum measured by the spectroscopic measurement device 10 in the absence of a measurement object is used as the reference spectroscopic spectrum.

The inspection device 11 performs preprocessing for a predetermined wavelength range of the absorption spectrum of tablet W. The predetermined wavelength range is received from the calibration curve generating device 2 connected via a communication port and stored in the storage device of the inspection device 11.

The preprocessing performed by the inspection device 11 is the same as the preprocessing performed by the preprocessing unit 20 of the calibration curve generation device 2, which will be described later, and therefore will not be described further. Once the preprocessing of the absorption spectrum of tablet W is complete, the inspection device 11 refers to the calibration curve stored in the storage device and calculates the measurement value corresponding to the absorption spectrum of tablet W.

In this embodiment, the measured value represents the content of the test target component contained in the tablet W. The calibration curve is received from the calibration curve generating device 2 connected via the communication port and stored in the memory device of the testing device 11.

The inspection device 11 determines whether the quality of the tablets W is good or bad based on the measured values. The inspection device 11 outputs a sorting signal based on the quality determination result to a sorting device (not shown) that sorts the tablets W into normal and defective products. The sorting device sorts the tablets W into normal and defective products based on the sorting signal.

The calibration curve generating device 2 generates a calibration curve that is referenced by the inspection device 11 to inspect the tablet W measured by the spectrometric measurement device 10. In this embodiment, the calibration curve generating device 2 is configured as a computer device equipped with a CPU, RAM, ROM, a storage device such as a hard disk drive, a communication port, a display device, and an input device.

The ROM and storage device of this computer device store a program that causes the computer device to function as the calibration curve generating device 2. In other words, the CPU uses the RAM as a working area to execute the program stored in the ROM and storage device, causing the computer device to function as the calibration curve generating device 2 in this embodiment.

The display device may be, for example, a liquid crystal display device. The input device may be, for example, a keyboard device or a pointing device. The input device may also be a touchpad integrated with the display device. A cable for communicating with the inspection device 11 is connected to the communication port.

The storage device of the calibration curve generating device 2 stores the absorption spectra of multiple tablets whose true values are known, associated with the true values. The true values of the multiple tablets are measured in advance using a device capable of high-precision measurements, such as a high-performance liquid chromatography device. The absorption spectra of the multiple tablets are measured in advance using the product inspection device 1.

In this embodiment, the absorption spectra of multiple tablets are grouped according to true values and stored in the storage device of the calibration curve generating device 2. For example, the first group is assigned the absorption spectra of 80 tablets with a true value near 2. The second group is assigned the absorption spectra of 80 tablets with a true value near 4. The third group is assigned the absorption spectra of 80 tablets with a true value near 8. The fourth group is assigned the absorption spectra of 80 tablets with a true value near 12.

The calibration curve generating device 2 includes a preprocessing unit 20, a classification unit 21, a calibration curve calculation unit 22, and a calibration curve evaluation unit 23.

The preprocessing unit 20 performs predetermined preprocessing on the absorption spectra stored in the storage device to convert them into a data format suitable for the multivariate analysis described below. The predetermined preprocessing includes at least one of the four arithmetic operations, differentiation, integration, and normalization.

The classification unit 21 classifies the preprocessed absorption spectra into teacher data Tr and test data Te. In this embodiment, the classification unit 21 classifies the preprocessed absorption spectra of 40 tablets from each of the first to fourth groups as teacher data Tr, and classifies the preprocessed absorption spectra of the remaining 40 tablets from each of the first to fourth groups as test data Te.

The classification unit 21 calculates statistical values of the first to fourth types for each preprocessed absorption spectrum. In this embodiment, the statistical value of the first type is the average value, the statistical value of the second type is the maximum value, the statistical value of the third type is the minimum value, and the statistical value of the fourth type is the standard deviation.

As shown in Figure 2, for each of the first to fourth groups, the classification unit 21 selects 10 absorption spectra to be classified into training data Tr from the 80 preprocessed absorption spectra so as to evenly cover the average values.

In addition, for each of the first to fourth groups, the classification unit 21 selects 10 absorption spectra to be classified as training data Tr from the 70 absorption spectra remaining, excluding the 10 absorption spectra classified as training data Tr from the 80 absorption spectra that have undergone preprocessing, so as to evenly cover the maximum values.

In addition, for each of the first to fourth groups, the classification unit 21 selects 10 absorption spectra to be classified as training data Tr from the 70 absorption spectra remaining, excluding the 20 absorption spectra classified as training data Tr from the 60 preprocessed absorption spectra, so as to evenly cover the minimum values.

Furthermore, for each of the first to fourth groups, the classification unit 21 selects 10 absorption spectra to be classified as training data Tr from the 50 absorption spectra remaining, excluding the 30 absorption spectra classified as training data Tr from the 80 preprocessed absorption spectra, so as to evenly cover the standard deviation.

In this way, for each of the first to fourth groups, the classification unit 21 selects 40 absorption spectra from the 80 preprocessed absorption spectra to classify as training data Tr, and classifies the remaining 40 absorption spectra as test data Te.

As shown in FIG. 3, the calibration curve calculation unit 22 calculates a calibration curve representing the correlation between the training data Tr and the true values by performing multivariate analysis on the training data Tr and the true values associated with the training data Tr. The calibration curve calculation unit 22 performs multivariate analysis using, for example, simple regression analysis, multiple regression analysis, quantification type 1, quantification type 2, quantification type 3, discriminant analysis, logistic regression analysis, principal component analysis, partial least squares regression (PLS regression), factor analysis, cluster analysis, correspondence analysis, multidimensional scaling, conjoint analysis, support vector machine, decision tree, random forest, naive Bayes, neural network, deep learning, etc.

In this embodiment, the calibration curve calculation unit 22 performs multivariate analysis using PLS regression. The calibration curve calculated by the calibration curve calculation unit 22 is transmitted to the inspection device 11 via a communication port in response to, for example, an operation of the input device, and is stored in the storage device of the inspection device 11.

As shown in FIG. 4, the calibration curve evaluation unit 23 calculates an evaluation value of the calibration curve calculated by the calibration curve calculation unit 22 using the test data Te and the true value associated with the test data Te, and evaluates the accuracy of the calibration curve based on the evaluation value.

In this embodiment, the calibration curve evaluation unit 23 calculates an evaluation value of the calibration curve based on at least one index, such as the mean square error, root mean square error, mean absolute error, mean error, prediction standard error, coefficient of determination, and coefficient of determination corrected for degrees of freedom. The calibration curve evaluation unit 23 displays the calculated evaluation value, for example, on a display device.

As described above, the calibration curve generation device 2 according to this embodiment improves the accuracy of the calibration curve by selecting absorption spectra to be classified as training data so as to provide high comprehensiveness for the statistical values obtained from each absorption spectrum, thereby preventing the inspection device 11 from making erroneous judgments about the inspection results.

In the embodiment described above, an example was described in which the first type of statistical value was the average value, the second type of statistical value was the maximum value, the third type of statistical value was the minimum value, and the fourth type of statistical value was the standard deviation.

In contrast, the first to fourth types of statistical values may be other combinations as long as the types do not overlap. Furthermore, each of the first to fourth types of statistical values may be an average value, maximum value, minimum value, standard deviation, or other types of statistical values such as a range (maximum value - minimum value). Furthermore, the classification unit 21 may allow the user to select the first to fourth types of statistical values via an input device of the calibration curve generating device 2.

Furthermore, as shown in FIG. 5, the classification unit 21 may display an image on the display device of the calibration curve generating device 2 to allow the operator to confirm that the training data Tr has been selected from a statistically sufficient number of absorption spectra so as to provide high comprehensiveness for the statistical values.

Figure 5 shows an example of an image display that allows the operator to confirm that 10 absorption spectra, numbered 1, 10, 19, 28, 37, 46, 55, 64, 73, and 80, in ascending order of average value, have been selected as training data Tr from the 80 absorption spectra.

Furthermore, when the inspection device 11 detects a tablet W that exceeds the tolerance range for the statistical values of the first to fourth types, it may send a sorting signal to sort the tablet W into a category different from normal and defective products, thereby prompting the calibration curve generating device 2 to regenerate the calibration curve.

In this case, when classifying the preprocessed absorption spectra into training data Tr and test data Te, the classification unit 21 calculates the tolerance ranges for the statistical values of the first to fourth types, and transmits the calculated tolerance ranges together with the calibration curve to the inspection device 11 via the communication port, where they are stored in the storage device of the inspection device 11.

Although embodiments of the present invention have been disclosed above, it will be apparent to those skilled in the art that modifications may be made without departing from the scope of the present invention. All such modifications and equivalents are intended to be encompassed by the following claims.

2 Calibration curve generating device 10 Spectroscopic measurement device 11 Inspection device 20 Pre-processing unit 21 Classification unit 22 Calibration curve calculation unit 23 Calibration curve evaluation unit

Claims (5)

A calibration curve generating device (2) that generates a calibration curve to be referenced by an inspection device (11) that inspects an article based on a spectroscopic spectrum of the article measured by a spectroscopic measurement device (10), comprising:
a pre-processing unit (20) that pre-processes a plurality of absorption spectra obtained by the spectroscopic measurement device for a plurality of articles whose true values are known;
a classification unit (21) that classifies the plurality of absorption spectra that have been subjected to the preprocessing into teacher data and test data;
a calibration curve calculation unit (22) for calculating the calibration curve from the teacher data and the true value;
a calibration curve evaluation unit (23) that evaluates the calibration curve from the test data and the true value,
The classification unit
Calculating statistics obtained from each of the pre-processed absorption spectra;
A calibration curve generating device that selects absorption spectra to be classified into the training data so as to cover all of the statistical values. the statistical values include a first type of statistical value and a second type of statistical value;
The classification unit
selecting absorption spectra to be classified into the training data so as to evenly cover the first type of statistical values;
2. The calibration curve generating device according to claim 1, wherein the absorption spectra to be classified as the training data are selected from the absorption spectra excluding the absorption spectra classified as the training data from the plurality of absorption spectra so as to evenly cover the second type of statistical values. The calibration curve generating device of claim 1, wherein the statistical value is one of the mean value, maximum value, minimum value, standard deviation, and range. A calibration curve generating method for generating a calibration curve to be referenced by an inspection device (11) that inspects an object based on a spectroscopic spectrum of the object measured by a spectroscopic measurement device (10), comprising:
a pre-processing step of pre-processing a plurality of absorption spectra obtained by the spectroscopic measurement device for a plurality of articles whose true values are known;
a classification step of classifying the plurality of preprocessed absorption spectra into training data and test data;
a calibration curve calculation step of calculating the calibration curve from the teacher data and the true value;
a calibration curve evaluation step of evaluating the calibration curve from the test data and the true value,
The classification step includes:
Calculating statistics obtained from each of the pre-processed absorption spectra;
A calibration curve generating method for selecting absorption spectra to be classified into the training data so as to cover all of the statistical values. A calibration curve generation program that causes a computer to generate a calibration curve to be referenced by an inspection device (11) that inspects an item based on a spectroscopic spectrum of the item measured by a spectroscopic measurement device (10), comprising:
a pre-processing step of pre-processing a plurality of absorption spectra obtained by the spectroscopic measurement device for a plurality of articles whose true values are known;
a classification step of classifying the plurality of preprocessed absorption spectra into training data and test data;
a calibration curve calculation step of calculating the calibration curve from the teacher data and the true value;
a calibration curve evaluation step of evaluating the calibration curve from the test data and the true value,
The classification step includes:
Calculating statistics obtained from each of the pre-processed absorption spectra;
a calibration curve generation program for selecting absorption spectra to be classified into the training data so as to cover all of the statistical values evenly;