Management method and management device for managing analyzer, and program

Abstract

To provide a management device that is connected to an analyzer, for example, a flow injection analyzer, using carrier fluid, and that detects occurrence of abnormality in a component, such as a pump, in the analyzer by discriminating the abnormality from other abnormalities.SOLUTION: A management device 6 comprises a processing section 61 generating a first discriminator 64 that categorizes and labels detection data of an analyzer 1 into normal data, first abnormality data being abnormality data due to abnormality in a prescribed component in the analyzer, and second data being abnormality data due to causes other than the abnormality in the prescribed component, and that conducts learning according to the labeled detection data, in order to discriminate the normal data, first abnormality data, and second abnormality data. After generating the first discriminator 65, the processing section 61 inputs newly outputted detection data to the first discriminator 65 in order to conduct discrimination of the detection data.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates to a method and device for managing analytical devices. [Background technology]

[0002] Some analytical devices for performing various chemical analyses inject a sample to be analyzed into a flow of a carrier fluid (carrier gas and carrier liquid), then pass the flow through a sample processing section that performs some operation or processing, and perform a detection operation on the fluid that has passed through the sample processing section. Since the detection value obtained by the detection operation changes over time, analytical devices that use the flow of a carrier fluid obtain detection data that is time-series data. Examples of devices that use the flow of a carrier fluid include liquid chromatography devices, gas chromatography devices, and flow injection analysis (FIA) devices. In liquid chromatography devices, after injecting a sample into a flow of carrier liquid, the carrier liquid is passed through a column that is a sample processing section to separate multiple components contained in the sample from each other, and a detector is used to measure, for example, absorbance, electrical conductivity, etc., on the liquid discharged from the column to obtain detection data. Similarly, a gas chromatography device injects a sample into a carrier gas flow, then flows the carrier gas through a column to separate multiple gas components contained in the sample from each other, and uses a detector to measure, for example, absorbance, thermal conductivity, etc., of the gas discharged from the column to obtain detection data. In an FIA device, a sample liquid is injected into a carrier liquid flow, and then a reagent liquid is added to the carrier liquid at a constant flow rate to react the analyte in the sample liquid with the reagent, and the reaction product is detected by measuring absorbance or electrical conductivity, etc., to obtain detection data. Patent Document 1 shows an example of an FIA device used to quantify urea in water.

[0003] In analytical devices that use fluid flow as described above, it is important that the flow of the carrier fluid is stable in order to obtain stable analytical results. When adding reagent liquid to the carrier liquid at a constant flow rate, such as in an FIA device, it is also important that the flow of the reagent liquid is stable. The carrier fluid and reagent liquid are delivered by a pump, but when this pump deteriorates, disturbances or pulsations can occur in the baseline of the time-series data from the detector. Until now, when disturbances occurred in the detection data or normal measurements could not be performed, maintenance on the pump was performed or the pump was replaced. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] JP 2018-179545 A Summary of the Invention [Problem to be solved by the invention]

[0005] In an analytical device that uses the flow of a carrier fluid, when normal measurement cannot be performed, the pump that supplies the carrier fluid is suspected to be deteriorated, and the pump is maintained or replaced. However, there are other reasons for not being able to perform normal measurement than pump deterioration. For example, in the case of a liquid chromatography device or FIA device, when air bubbles are contained in the liquid discharged from the sample processing unit and the air bubbles pass through the detector, spike-like noise occurs in the detection data. Air bubbles are not necessarily generated by pump deterioration, but can also occur when the liquid supplied to the pump itself contains air bubbles or when dissolved gas foams. Therefore, replacing a pump based only on the fact that normal measurement cannot be performed includes the possibility of replacing a pump that is not deteriorated, which is not economical. In addition, when other parts, such as an ultraviolet lamp used as a light source, are replaced at the same time as the pump is replaced, unnecessary replacement of parts such as the ultraviolet lamp occurs when a pump that is not deteriorated is replaced. It is also possible to predict the maintenance timing and lifespan of the pump, i.e., to assess the deterioration risk, based on the frequency of occurrence of the event that the measurement could not be performed normally. However, if the events that the measurement could not be performed normally include events that are unrelated to pump deterioration, it is not possible to accurately predict the maintenance timing and lifespan. It is also desirable to detect the occurrence of abnormalities in components other than the pump that constitute the analytical device based on the detection data obtained by the analytical device, and to assess the deterioration risk. Even in an analytical device that does not use the flow of a carrier fluid, it is desirable to accurately predict the maintenance timing and lifespan of the components that constitute the analytical device.

[0006] An object of the present invention is to provide a management method in an analytical device that can detect an abnormality in a specific component, such as a pump, by isolating it from other abnormalities, and a management device and program used to implement such a management method. [Means for solving the problem]

[0007] A management method of one embodiment of the present invention is a management method for managing an analytical device having a detection unit that performs detection operations and outputs detection data, which is time-series data, and classifies and labels the detection data into normal data, first abnormal data, which is abnormal data caused by an abnormality in a specified component of the analytical device, and second data, which is abnormal data caused by a cause other than the abnormality of the specified component, and performs learning using the labeled detection data to generate a first discriminator that discriminates between normal data, the first abnormal data, and the second abnormal data.After generating the first discriminator, new detection data output from the detection unit is input to the first discriminator to discriminate the detection data.

[0008] A management device of one embodiment of the present invention is a management device that manages an analytical device having a detection unit that performs detection operations and outputs detection data that is time-series data, and is equipped with a processing unit that acquires the detection data, classifies and labels the detection data into normal data, first abnormal data, which is abnormal data caused by an abnormality in a specified component of the analytical device, and second data, which is abnormal data caused by a cause other than the abnormality of the specified component, and learns using the labeled detection data to generate a first discriminator that discriminates between normal data, the first abnormal data, and the second abnormal data.After generating the first discriminator, the processing unit inputs new detection data output from the detection unit to the first discriminator to discriminate the detection data.

[0009] Another aspect of the management device of the present invention is a management device that manages an analytical device having a detection unit that performs detection operations and outputs detection data, which is time-series data, and has a first discriminator that discriminates the detection data into normal data, first abnormal data, which is abnormal data caused by an abnormality in a specified component of the analytical device, and abnormal data caused by a cause other than the abnormality of the specified component, and a processing unit that inputs the detection data to the first discriminator and obtains a discrimination result from the first discriminator.

[0010] A program according to one embodiment of the present invention causes a computer connected to an analytical device having a detection unit that performs detection operations and outputs detection data, which is time-series data, to perform a process of learning based on detection data that has been classified and labeled into normal data, first abnormal data, which is abnormal data caused by an abnormality in a specified component in the analytical device, and second data, which is abnormal data caused by a cause other than the abnormality in the specified component, to generate a first discriminator that discriminates whether the detection data is normal data, the first abnormal data, or the second abnormal data, and a process of inputting newly output detection data from a detector into the first discriminator after the generation of the first discriminator to discriminate the detection data. Effect of the Invention

[0011] According to the present invention, in an analytical device, it becomes possible to detect an abnormality occurring in a specific component such as a pump while separating it from other abnormalities. [Brief description of the drawings]

[0012] [Figure 1] FIG. 2 is a block diagram showing a system including an analysis device and a management device. [Diagram 2] FIG. 1 is a block diagram showing a system including a flow injection analysis apparatus and a management apparatus. [Diagram 3] Graphs (a) to (f) show examples of detection data that is time-series data. [Figure 4] 13 is a flowchart showing a process of detecting the occurrence of an abnormality and evaluating a deterioration risk. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0013] Next, an embodiment of the present invention will be described with reference to the drawings. Fig. 1 shows a system consisting of an analysis device 1 and a management device 6. The analysis device 1 is an analysis device that utilizes the flow of a carrier fluid, and the management device 6 is a device used to manage the analysis device 1, and detects the occurrence of an abnormality in the analysis device 1 based on detection data obtained from the analysis device 1, determines the type of the abnormality, and further predicts the occurrence of the abnormality.

[0014] The analyzer 1 is provided with a pipe 2, one end of which is connected to a sample processor 4 and supplies a carrier fluid to the sample processor 4. The pipe 2 is provided with a pump P1 for forming a flow of the carrier fluid, and an injection section 3 provided between the secondary side, i.e., the outlet, of the pump P1 and the sample processor 4. A sample to be analyzed in the analyzer 1 is supplied to the injection section 3, and the injection section 3 injects a sample into the flow of the carrier fluid. The flow of the carrier fluid into which the sample has been injected passes through the sample processor 4, and undergoes some processing or operation while passing through the sample processor 4. When the analyzer 1 is configured as a liquid chromatography device or a gas chromatography device, the sample processor 4 is a chromatography column that is a stationary phase. When the analyzer 1 is configured as a flow injection analysis (FIA) device, the sample processor 4 is a mixing section for reaction reagents and a reaction coil provided downstream of the mixing section. A detection section 5 is provided downstream of the sample processor 4, and the fluid that has passed through the sample processor 4 is supplied to the detection section 5. The detection unit 5 is configured with an absorbance sensor, a thermal conductivity sensor, an electrical conductivity sensor, etc., and outputs detection values ​​that change from moment to moment. Time-series data of the detection values ​​is called detection data.

[0015] The management device 6 is configured as a general computer device, and includes a processing unit 61 that inputs time-series data from the detection unit 5, a storage unit 62 that stores data, and a display unit 63 that displays the results obtained by the management device 6. In the management device 6, a first discriminator 64 and a second discriminator 65 are generated by performing machine learning. These discriminators 64 and 65 are constructed as machine learning models and output a discrimination result when data is input. As an example, the discriminators 64 and 65 calculate a plurality of feature amounts from the input data, classify the data according to a combination of the values ​​of the feature amounts, and output a discrimination result. Such discriminators 64 and 65 can be expressed as a set of parameters that specify the calculation and classification of the feature amounts, and these parameters are generally stored in the storage unit 62, but these parameters may be stored in a storage area prepared separately from the storage unit 62 that is provided as a general-purpose item in the management device 6. Generating these parameters by machine learning and determining their values ​​is the generation of the discriminators 64 and 65. From the viewpoint that the parameters are stored in the storage unit 62, it can also be said that the discriminators 64 and 65 are stored in the storage unit 62. The process of generating the discriminators 64 and 65 can be executed by the processing unit 61, but a processing device that executes the process of generating the discriminators 64 and 65 may be provided separately from the processing unit 61. The storage area that stores the discriminators 64 and 65 and the processing device that generates the discriminators 64 and 65 may be provided away from the processing unit 61. The processing device that generates the discriminators 64 and 65 is also configured as a general computer device. Note that the discriminators 64 and 65 can also be provided as independent hardware devices by incorporating a machine learning model into a computer device independent of the processing unit 61. In that case, generating the machine learning model to be incorporated into the computer device and incorporating it into the computer device is to generate the discriminators 64 and 65.

[0016] In order to improve the analytical accuracy in the analytical device 1 shown in Fig. 1, it is necessary that the flow of the carrier fluid is stable, and therefore it is necessary that the pump P1 operates normally. If the pump P1 deteriorates and its operation becomes unstable, the flow of the carrier fluid also becomes unstable, and the baseline of the detection data may fluctuate significantly or the baseline may pulsate. As a result, the peak due to the detection target substance may be buried in the baseline fluctuation, making it impossible to perform normal measurement or reducing the quantitative accuracy.

[0017] Next, the operation of the management device 6 in the system shown in FIG. 1 will be described. The management device 6 analyzes the detection data from the analysis device 1, detects the occurrence of an abnormality such as deterioration in components such as a pump constituting the analysis device 1, and further performs prediction regarding the occurrence of an abnormality in the components, that is, deterioration risk evaluation. For this purpose, the management device 6 first creates a model for classifying whether a peak included in the detection data is a peak due to normal measurement or a peak due to an abnormality in the measurement. This model is created as a first discriminator 64 and incorporated into the management device 6. In creating the model, detection data that is known to be normal data and detection data that is known to be abnormal data are used to label the detection data, and machine learning is performed using the labeled detection data. At this time, classification is not performed by simply labeling the detection data according to normal detection data and abnormal detection data, but abnormal data is classified and labeled as abnormal data (first abnormal data) caused by deterioration of a specific component or abnormal data (second abnormal data) caused by a specific reason other than the deterioration of a specific component.

[0018] Once the first discriminator 64 is created, the detection data can be classified using the first discriminator 64. The management device 6 then inputs the detection data (referred to as unknown data) for which it is unclear whether it is normal data or abnormal data to the first discriminator 64 to classify the unknown data. By repeatedly classifying the unknown data and accumulating the data, it is possible to obtain data (referred to as maintenance data) showing the relationship between the operating time of a component, for example a pump, and the frequency of occurrence of abnormal peaks due to deterioration of the component. The management device 6 accumulates the maintenance data and performs machine learning to create a deterioration risk assessment model that gives a deterioration risk assessment result for the component. The deterioration risk assessment model is created as the second discriminator 65 and incorporated into the management device 6. The deterioration risk assessment includes predicting the time when an abnormality will occur in the component, the time when the component will be replaced, the frequency of replacement, etc. Once the second discriminator 65 is created, the detection data classified as abnormal data by the first discriminator 64 is input to the second discriminator 65, and the second discriminator 65 outputs a deterioration risk assessment result for the component. The deterioration risk assessment result thus obtained is displayed on the display unit 63. The second discriminator 65 may generate an alarm when the occurrence frequency of abnormal data exceeds a certain threshold. The threshold is set based on the result of machine learning.

[0019] In FIG. 1, the management device 6 is provided close to the analysis device 1, but the management device 6 may be provided at a location remote from the analysis device 1 and configured to receive detection data transmitted from the analysis device 1, or may be connected to the analysis device 1 via a network such as the Internet. Furthermore, it is also possible to operate the management device 6 in a so-called cloud form by having a server on the network function as the management device 6. In the cloud form, the discriminators 64, 65 and a processing device that executes a process to generate the discriminators 64, 65 are provided on a server on the network, and the detection data can be transmitted to the server from a processing unit 61 that receives detection data from the analysis device 1. In the cloud form, if there are multiple analysis devices 1 of the same or similar model or configuration, the discriminators 64, 65 may be provided in common to the multiple analysis devices 1, or when the discriminators 64, 65 are generated and no further learning is required, the discriminators 64, 65 can be individually incorporated into the management device 6 connected to each analysis device 1. In other words, even if the management device 6 connected to the analysis device 1 does not itself have the function of generating the discriminators 64, 65, if the learned discriminators 64, 65 are incorporated in the management device 6, the management device 6 is included in the category of the management device of the present invention.

[0020] As a machine learning method for generating the classifiers 64 and 65, for example, a random forest can be used, but other methods can also be used. It is also possible to generate the classifiers 64 and 65 using a neural network. Since using a neural network increases the computation load, it is preferable to use a machine learning method such as a random forest to generate the classifiers 64 and 65.

[0021] The present invention will be described in more detail below based on specific embodiments. Fig. 2 shows a specific configuration of a system including an analysis device 1 and a management device 6 when the analysis device 1 is an FIA device 10.

[0022] First, the FIA ​​device 10 will be described. A six-way valve 11 is provided as an injection section 3 for injecting sample liquid into the flow of carrier liquid. The six-way valve 11 is generally used in FIA devices and liquid chromatography devices for sampling samples, that is, for injecting a certain volume of sample liquid into the flow of carrier liquid, and has six ports A to F. A pipe 21 to which carrier liquid is supplied via a pump P1 is connected to port F of the six-way valve 11, and a pipe 22 to which sample liquid is supplied via a pump P2 is connected to port B. A pipe 25 for draining sample liquid from the six-way valve 11 via a pump P4 is connected to port C. A sample loop 12 for collecting a predetermined volume of sample liquid is connected between ports A and D. One end of a pipe 24 serving as an outlet of the six-way valve 11 is connected to port E.

[0023] The other end of the pipe 24 is connected to one end of the reaction coil 31, and the pipe 24 is provided with a mixing section 43 that mixes the reagent liquid with the flow of the carrier liquid at a predetermined flow rate. The mixing section 43 is connected to a pipe 26 to which the reagent liquid is supplied via a pump P3. The reagent liquid contains a reaction reagent that reacts with a target component in the sample liquid at a predetermined concentration. A detector 32 is provided downstream of the reaction coil 31, and the flow that has passed through the reaction coil 31 is supplied to the detector 32. The detector 32 corresponds to the detection section 5 in the analysis device 1 shown in FIG. 1, and is, for example, an absorbance sensor or an electrical conductivity sensor, and detects a product of the reaction between the target component in the sample and the reaction reagent in the reagent liquid. A back pressure coil 33 is provided downstream of the detector 32, and the liquid that has passed through the detector 32 is discharged from the outlet of the back pressure coil 33. The detection data from the detector 32 is supplied to a management device 6. The management device 6 used in the system shown in FIG. 2 is the same as that shown in FIG. 1.

[0024] In the hexagonal valve 11, the communication between port X and port Y is represented as (XY). The hexagonal valve 11 can be switched between a first state of (AB), (CD), and (EF) and a second state of (BC), (DE), and (FA). In FIG. 2, the connection between the ports in the first state is shown by solid lines, and the connection between the ports in the second state is shown by dotted lines. In the first state, the carrier liquid is pumped by the pump P1 and flows from the pipe 21 to port F to port E to pipe 24, and flows out of the hexagonal valve 11 to the downstream side. The sample liquid flows from the pipe 22 to port B to port A to the sample loop 12 to port D to port C, and is discharged from the pipe 25. When the first state is switched to the second state, the sample liquid flows from the pipe 22 to port B to port C and is discharged from the pipe 25, and the carrier liquid flows from the pipe 21 to port F to port A to the sample loop 12 to port D to port E to the pipe 24 and flows out to the downstream side. At this time, the sample liquid that has already flowed in and filled the sample loop 12 when the first state was in effect flows from port E into the pipe 24 before the carrier liquid, and flows to the downstream side of the six-way valve 10. The volume of the sample liquid flowing into the pipe 24 is determined by the sample loop 12. Therefore, by repeatedly switching between the first state and the second state (for example, by rotating the six-way valve 11 in the direction of the arrow in the figure), a predetermined volume of sample liquid can be repeatedly sent into the pipe 24.

[0025] When a predetermined amount of sample liquid is injected into the flow of carrier liquid by operating the six-way valve 11, the reagent liquid is mixed with the sample liquid at a constant flow rate by the mixer 43, the target component reacts with the reaction reagent in the reaction coil 31, and the reaction product is detected by the detector 32. The target component in the sample liquid can be quantified based on the peak intensity (peak height) or the peak area (i.e., integrated intensity) in the detection value of the detector 32. The six-way valve 11 can be switched between the first state and the second state at predetermined time intervals, taking into account the residence time required for the reaction and the time required for detection by the detector 32. In addition, the switching can also be performed by detecting that the sample liquid introduced into the detector 32 has been discharged from the detector 32. In this way, the switching between the first state and the second state can be automatically performed, thereby enabling continuous measurement.

[0026] In the FIA ​​device 10 shown in FIG. 2, when the flow rate of the reagent liquid fluctuates, the flow of the carrier liquid in the reaction coil 31 and the detector 32 also fluctuates. Therefore, in order to perform a normal measurement, not only the pump P1 that feeds the carrier liquid but also the pump P3 that feeds the reagent liquid must be operating normally. In the FIA ​​device 10, abnormal data also occurs when air bubbles pass through the detector 32. Therefore, the management device 6 in the system shown in FIG. 2 classifies the detection data, which is time-series data from the detector 32, into three types: normal data, abnormal data due to deterioration of the pumps P1 and P3 (called pump deterioration abnormal data), and abnormal data due to air bubbles passing through the detector 32 (called air bubble abnormal data), to detect the occurrence of an abnormality, and further predicts the occurrence of an abnormality in the pumps P1 and P3 by accumulating and learning the abnormal data. In the example shown here, the pump deterioration abnormal data corresponds to the first abnormal data, and the air bubble abnormal data corresponds to the second abnormal data.

[0027] 3(a) to (f) show examples of detection data. FIG. 3(a) shows detection data including an ideal peak when the detector 32 detects the detection target substance as normal data. FIG. 3(b) shows detection data including a peak of the detection target substance, and includes noise slightly larger than that shown in FIG. 3(a). The detection data shown in FIG. 3(b) can also accurately quantify the detection target substance, so this detection peak is a normal peak. FIG. 3(c) shows an example of pump deterioration abnormal data. In this case, the baseline fluctuation in the detection data is large, the noise is large, and a peak due to pulsation is present. From such detection data, it is difficult to determine whether or not a peak of the detection target substance exists, and even if it is determined that a peak of the detection target substance exists, it is difficult to accurately quantify the detection target substance. FIG. 3(d) to (f) show examples of bubble abnormal data. The bubble abnormal peak is characterized by having a spike-like peak that occurs when bubbles pass through the detector 32 and is significantly narrower than the peak due to the detection of the detection target substance. FIG. 3(d) shows bubble abnormality data in a state where there is no peak due to the detection target substance, and FIG. 3(e) and (f) show bubble abnormality data in which a spike-like peak due to bubbles is superimposed on a peak due to the detection target substance. Comparing FIG. 3(e) and FIG. 3(f), the detection data in FIG. 3(f) has more noise, but in both cases, the peak due to the detection target substance and the spike-like peak due to bubbles can be separated by calculation, and therefore, it is possible to quantify the detection target substance. Although not shown in FIG. 3, when a discharge lamp is used as a light source and the light source deteriorates, a step-like noise may occur. Therefore, from these examples of detection data, it can be seen that, for example, pump deterioration abnormality data, bubble abnormality data, and abnormality data caused by deterioration of the discharge lamp, which is the light source, have different shapes, etc. Therefore, it can be seen that these can be clearly distinguished by a discriminator created by machine learning.

[0028] The flowchart shown in FIG. 4 explains the process of detecting the occurrence of an abnormality and evaluating the deterioration risk in the system shown in FIG. 2. In the initial state, it is assumed that neither the first discriminator 64 nor the second discriminator 65 has been created. First, in step 101, detection data, which is time-series data, is acquired from the detector 32, and in step 102, it is determined whether the first discriminator 64 has been created. At this stage, the first discriminator 64 has not been created, so the process proceeds to step 103. In step 103, the detection data is stored in the memory unit 62, and in step 104, labeling is performed on the detection data stored in the memory unit 62. In the labeling, one of three types of labels, normal data, pump deterioration abnormality data, and air bubble abnormality data, is assigned to each piece of detection data. Thereafter, in step 105, , Inspection A learning model for classifying the detected data is created as a first discriminator 64 by machine learning. The learning data used for machine learning uses feature quantities extracted from the labeled detection data. Examples of feature quantities that can be used include statistics such as maximum, minimum and median values ​​in the detection data, the number, height and position of peaks, and frequency dependence (frequency spectrum) of data obtained by Fourier transforming the detection data. Once the first discriminator 64 is created, by inputting the detection data to it, it becomes possible to detect the occurrence of pump deterioration anomalies independently of anomalies caused by air bubbles.

[0029] After the first discriminator 64 is created, the process from step 101 is repeated, and since the first discriminator 64 has already been created, the process proceeds from step 102 to step 111. In step 111, the detection data is input to the first discriminator 64, and the detection data is discriminated as normal data, pump deterioration abnormality data, or air bubble abnormality data. Furthermore, in step 112, the discriminated pump deterioration abnormality data is subjected to normal value analysis, abnormality detection history recording, peak cutting processing, and the like, and the pump deterioration abnormality data subjected to these processes is stored in the storage unit 62 in step 113. Next, in step 114, it is determined whether the second discriminator 65 has already been created. Since the second discriminator 65 has not yet been created at this stage, the process proceeds to step 115. In step 115, a pump deterioration risk assessment model is created as the second discriminator 65 by machine learning using the data stored in the storage unit 52 in step 113. The pump deterioration risk assessment model is a learning model for predicting at least one of the timing of pump deterioration and the timing of pump replacement, and the features used to create the pump deterioration risk assessment model include the pump deterioration anomaly data processed in step 112, as well as the frequency of occurrence of pump deterioration anomalies divided into specific time periods (i.e., the frequency of occurrence of pump deterioration anomalies), the operating times of pumps P1 and P3, etc.

[0030] After the first discriminator 64 and the second discriminator 65 are created, the process from step 101 is repeated again, and the process proceeds from step 102 to step 111, where the first discriminator 64 judges the normality of the detection data. Thereafter, the process proceeds from step 114 to step 121, where the second discriminator 65 evaluates the risk of pump deterioration. The result is output to the memory unit 62 or the display unit 63 as a judgment result on the possibility of pump deterioration and a prediction result for the timing of pump replacement, and the series of processes is then completed.

[0031] Thereafter, the process shown in FIG. 4 is executed each time the FIA ​​device 10 quantifies the detection target substance. In this case, since the discriminators 64 and 65 have already been created, the process does not proceed from step 102 to step 103 or from step 114 to step 115, and the pump deterioration risk is evaluated each time detection data is acquired in step 101. In this embodiment, it is possible to improve the first discriminator 64 by further performing machine learning after the first discriminator 64 is created. When improving the first discriminator 64, when proceeding from step 102 to step 111, the process of step 103 (storing in the memory unit 62) and step 104 (labeling) are also executed, and the already created first discriminator 64 is re-learned. Furthermore, when there are multiple FIA ​​devices 10 of the same or similar configuration or model, the discriminators 64, 65 are created using one of the FIA ​​devices 10, and then the discriminators 64, 65 are incorporated into the management device 6 connected to the other FIA devices 10. This makes it possible to obtain a judgment result regarding the possibility of deterioration of the pumps P1, P3 in the other FIA devices 10 and a prediction result of the timing for pump replacement each time detection data is obtained in the other FIA devices 10.

[0032] In the above description, when the specific component is a pump, when classifying abnormal data among the detection data into pump deterioration abnormal data and abnormal data due to causes other than pump deterioration, air bubble abnormal data is used as abnormal data due to causes other than pump deterioration. In addition to air bubble abnormal data, usable abnormal data generated due to causes other than pump deterioration include abnormal data in which noise is increased due to dirt in the detector 32 or deterioration of the light source attached to the detector 32, abnormal data in which the baseline fluctuates in a sawtooth shape due to bubbles remaining in the detector 32, abnormal data with continuous columnar noise due to the precipitation of solid matter in the detector 32, and abnormal data in which the baseline changes in a meandering manner due to changes in the ambient temperature. If the abnormal data generated due to causes other than pump deterioration can be determined to be separate from abnormal data generated due to abnormalities or deterioration in the pumps P1 and P3, these abnormal data can be used instead of air bubble abnormal data.

[0033] In the above-described embodiment, the specified component in the analysis device is a pump, and abnormal peaks caused by abnormalities or deterioration of the pump are identified, and further, from the occurrence frequency of the abnormal peaks thus identified, the time of occurrence of deterioration in the pump and the time of replacement of the pump are predicted. However, the components to which the present invention can be applied are not limited to pumps. For example, when the analysis device includes a light source, by using a classifier that has been trained to be able to identify abnormal peaks caused by deterioration of the light source, it is possible to identify abnormal peaks caused by deterioration of the light source from among abnormal peaks, and to predict the time of deterioration and the time of replacement of the light source. Furthermore, in the present invention, the specific component, i.e., the component that is the target of detection of abnormalities such as deterioration, is not limited to one. For example, multiple components such as a pump and a discharge lamp used as a light source can be specified as specific components. In this case, for example, the characteristics shown by detection data when an abnormality occurs are different between a pump and a discharge lamp, so when abnormal data caused by deterioration of a specific component is identified and labeled, different labels can be used depending on which component the abnormal data is caused by deterioration in. This makes it possible to simultaneously detect deterioration in multiple components and simultaneously perform deterioration risk evaluation. [Explanation of symbols]

[0034] 1 Analyzer 3 Injection part 4. Sample Processing Section 5. Detection section 6 Management device 10 Flow Injection Analysis (FIA) Apparatus 11 Six-way valve 30 Reaction Thermostat 31 Reaction coil 32 Detector 33 Back pressure coil 43 Mixing section 61 Processing section 62 Storage section 53 Display section 64,65 Discriminator

Claims

1. A management method for managing an analytical device having a detection unit that performs a detection operation and outputs detection data that is time-series data, comprising the steps of: classifying and labeling the detection data into normal data, first abnormal data which is abnormal data caused by an abnormality in a predetermined component of the analysis device, and second data which is abnormal data caused by a cause other than the abnormality of the predetermined component; generating a first classifier that discriminates between the normal data, the first abnormal data, and the second abnormal data by performing learning using the labeled detection data; a management method including inputting the detection data newly output from the detection unit to the first discriminator after the first discriminator is generated, and discriminating the detection data.

2. After generating the first classifier, a second classifier is created by performing learning using at least the operating time of the predetermined component and the occurrence frequency of the first abnormal data determined by the first classifier, and evaluating a deterioration risk of the predetermined component; The management method according to claim 1, wherein after the second discriminator is generated, the detection data that is input to the first discriminator and discriminated to be the first abnormal data is input to the second discriminator, thereby obtaining a prediction result regarding deterioration of the specified component as a discrimination result.

3. The management method according to claim 1 or 2, wherein the analytical device includes a sample processing unit that processes a flow of carrier fluid into which a sample has been injected, and the detection unit is provided downstream of the sample processing unit and performs the detection operation on the flow that has passed through the sample processing unit.

4. The management method according to claim 3, wherein the carrier fluid is a carrier liquid, the specified component includes at least a pump used to supply the carrier liquid, and the second abnormal data is abnormal data that occurs in the detection data when an air bubble passes through the detection section.

5. 5. The management method according to claim 4, wherein the analytical device is a flow injection analytical device that utilizes a reaction between the sample and a reagent solution, and the predetermined components further include a pump used for feeding the reagent solution.

6. A management device that manages an analysis device having a detection unit that performs a detection operation and outputs detection data that is time-series data, a processing unit that acquires the detection data, classifies and labels the detection data into normal data, first abnormal data which is abnormal data caused by an abnormality in a predetermined component of the analysis device, and second abnormal data which is abnormal data caused by a cause other than the abnormality of the predetermined component, and generates a first discriminator that discriminates between the normal data, the first abnormal data, and the second abnormal data by learning using the labeled detection data, The processing unit, after generating the first discriminator, inputs the detection data newly output from the detection unit to the first discriminator to discriminate the detection data.

7. The management device of claim 6, wherein the processing unit, after generating the first discriminator, performs learning using at least the operating time of the specified component and the frequency of occurrence of the first abnormal data to create a second discriminator that evaluates the risk of deterioration, and after generating the second discriminator, inputs the detection data that has been input to the first discriminator and determined to be the first abnormal data into the second discriminator, thereby obtaining a deterioration risk evaluation result for the specified component as a discrimination result.

8. A management device that manages an analysis device having a detection unit that performs a detection operation and outputs detection data that is time-series data, a first discriminator that discriminates the detection data into normal data, first abnormal data which is abnormal data caused by an abnormality in a predetermined component of the analysis device, and second abnormal data which is abnormal data caused by a cause other than the abnormality of the predetermined component; a processing unit that inputs the detection data to the first discriminator and obtains a discrimination result from the first discriminator; A management device having the above configuration.

9. a second discriminator that performs a deterioration risk assessment for the predetermined component based on at least an operating time of the predetermined component and an occurrence frequency of the first abnormality data discriminated by the first discriminator; The management device according to claim 8 , wherein the processing unit inputs the first abnormality data discriminated by the first discriminator to the second discriminator to obtain a deterioration risk evaluation result.

10. The management device according to any one of claims 6 to 9, wherein the analytical device includes a sample processing unit that processes a flow of carrier fluid into which a sample has been injected, and the detection unit is provided downstream of the sample processing unit and performs the detection operation on the flow that has passed through the sample processing unit.

11. The management device of claim 10, wherein the carrier fluid is a carrier liquid, the specified components include at least a pump used to supply the carrier liquid, and the second abnormal data is abnormal data that occurs in the detection data when an air bubble passes through the detection section.

12. 12. The management device of claim 11, wherein the analytical device is a flow injection analytical device that utilizes a reaction between the sample and a reagent solution, and the predetermined components further include a pump used to feed the reagent solution.

13. A computer connected to an analysis device having a detection unit that performs a detection operation and outputs detection data that is time-series data, a process of performing learning based on the detection data classified and labeled into normal data, first abnormal data which is abnormal data caused by an abnormality in a specific component of the analysis device, and second data which is abnormal data caused by a cause other than the abnormality of the specific component, and generating a first discriminator that discriminates whether the detection data is the normal data, the first abnormal data, or the second abnormal data; a process of inputting the detection data newly output from the detection unit to the first discriminator after generating the first discriminator, and causing the first discriminator to discriminate the detection data; A program that executes the following.

14. The computer further comprises: After generating the first discriminator, a process of performing learning using at least the operating time of the predetermined component and the occurrence frequency of the first abnormal data discriminated by the first discriminator, and creating a second discriminator that gives a deterioration risk evaluation result of the predetermined component; a process of obtaining the deterioration risk evaluation result by inputting the detection data determined to be the first abnormal data by the first discriminator to the second discriminator after generating the second discriminator; The program according to claim 13,

Images