VECTOR-BASED SIMILARITY ANALYSIS FOR AN ANALYSIS DEVICE

Abstract

An analysis device (10), in particular a sample separation device, for carrying out an analysis is described, which has a control device (100) which is designed to: i) determining at least one operating parameter relating to the analysis device (10); ii) creating at least two vectors (110, 120) relating to the at least one operating parameter; and iii) performing a vector-based similarity analysis based on the at least two created vectors (110, 120).

Description

FIELD OF THE INVENTIONThe present invention relates to an analysis device for carrying out an analysis, which has a control device which is configured to: determine at least one operating parameter with respect to the analysis device, generate at least two vectors with respect to the at least one operating parameter, and carry out a vector-based similarity analysis based on the at least two generated vectors. The invention further relates to a computer-implemented method for operating an analysis device, and to a device for data processing which can execute the method.BACKGROUND ARTAnalysis devices such as sample separators are provided for the analysis of a sample, in particular fluidic sample, e.g. for carrying out a chromatographic separation of the sample. For example, in a high performance liquid chromatography (HPLC) analysis device, a liquid (mobile phase) is moved through a so-called stationary phase (for example in a chromatographic column) at a very accurately controlled flow rate (for example in the range of microliters to milliliters per minute) and at a high pressure (typically 20 to 1000 bar and beyond, at present, up to 2000 bar) at which the compressibility of the liquid can be perceptible, in order to separate individual fractions of a sample liquid introduced into the mobile phase from one another. After passing through the stationary phase, the separated fractions of the fluidic sample are detected in a detector. Such an HPLC system is known, for example, from EP 0,309,596 B1 of the same applicant, Agilent Technologies, Inc.Such an analysis device may be very complex in operation and a plurality of operating parameters (many of which are directly dependent on each other) must always be monitored to ensure reliable operation. Analysis devices can therefore be equipped with a multiplicity of sensors in order to continuously collect data or operating parameters during the operation of the analysis device, which data or operating parameters then have to be correspondingly evaluated.Conventional solutions for the data analysis of instrument signals generally operate with defined reference signals and / or threshold values, with which the measured data are compared. However, during the lifetime of an instrument, these defined references / thresholds must be updated or adjusted periodically. Moreover, the references must be separately defined for each operating state of the device (e.g., for each chromatographic method, for each set of device parameters, etc.). For an operator of the instrument, it may be challenging to keep the overview here. Accordingly, it can quite happen that relevant deviations from operating parameters are missed in the provided data flood.DISCLOSUREThere may be a need to operate an analysis device efficiently and reliably. The object is achieved by means of the independent claims. Further embodiments are shown in the dependent claims.According to a first exemplary embodiment of the present invention, an analysis device is described (in particular a sample separation device such as, for example, an HPLC) for carrying out an analysis (for example separating a fluidic sample), (wherein the analysis device has a control device which is configured) for i) determining at least one operating parameter (for example viewed over time and / or a plurality of operating parameters) with respect to the analysis device. ii) creating at least two vectors (which can be compared, for example an actual state vector and a setpoint state vector of the operating parameter) with respect to the at least one operating parameter. iii) carrying out a vector-based similarity analysis based on the at least two created vectors.According to a second exemplary embodiment of the present invention, a (computer-implemented method) for operating an analysis apparatus (in particular a sample separation device) is described, the method having: i) determining at least one operating parameter with respect to the analysis apparatus; ii) creating at least two vectors with respect to the at least one operating parameter; iii) carrying out a vector-based similarity analysis based on the at least two created vectors.According to a third embodiment of the present invention, a device for data processing (e.g. one or more processors) is described which is configured to carry out the method as described above. The data processing apparatus may comprise one or more processors in a simple example. In a more complex example, the device for data processing can be a control device, in particular a central control device, of the analysis device. In one example, the control device may include central control software.According to a fourth embodiment of the present invention, use of vector-based similarity analysis for recognizing an abnormality regarding the operation of an analysis apparatus will be described.In the context of the present document, the term "operating parameter" is understood to mean, in particular, a measurable variable (measured variable, measured value) which at least partially describes a state of a system (during operation). An example of such an operating parameter, in particular in the context of an analysis device such as an HPLC, is e.g. the pressure, in particular the system pressure or the pressure at any point in a flow path. Another operating parameter is, for example, the temperature, the flow rate, or a detector signal. Operating parameters can be independent of one another or can be connected to one another (inextricably). In an exemplary embodiment, increasing the temperature in a column furnace will result in the flow rate increasing or the back pressure decreasing. In one (such) exemplary embodiment, an operating parameter allows conclusions to be drawn about another operating parameter. In one exemplary embodiment, an operating parameter can have a control variable and / or a controlled variable. Operating parameters can relate not only to internal factors (of the analysis device) but also to external factors, for example ambient temperature, atmospheric humidity, or pH value of an operating liquid. In one exemplary embodiment, the detection of an operating parameter enables a conclusion to be drawn about the operating state or at least a partial aspect of the operating state of the system. In one embodiment, one or more operating parameters may be considered over time. In a further exemplary embodiment, a plurality of operating parameters (dependent on or independent of one another) can be considered.In the context of the present document, the term "vector" can be understood in particular to mean a (mathematical) variable which is defined by a direction and an amount (length). A vector as such is known from mathematics and may be represented in a coordinate system (in particular as an arrow) and may be added, multiplied or compared with other vectors, for example. A vector can in principle have an infinite number of dimensions.In the present context, the term "creating a vector" can refer in particular to at least one operating parameter being brought into a vector form. Accordingly, the at least one operating parameter can be represented in the form of an arrow in a coordinate system. If at least two vectors are created based on operating parameters, these can be compared with one another, for example with regard to their position in the coordinate system, their length and / or their direction. In an exemplary embodiment, the vector may be represented in an n-dimensional space, e.g., with each dimension corresponding to the pressure at a particular time. In a further exemplary embodiment, for example, a vector for the pressure and a further vector for the temperature can be created (in particular in each case as a time profile). These vectors may be combined into a single vector, e.g., via addition or multiplication. In another embodiment, each dimension of a vector may correspond to a particular operating parameter. In one embodiment, a data matrix may also be created. In the present context, such a data matrix can be understood as an extended vector or a specific vector form. Therefore, the term "vector" in the present context may also comprise a data matrix. In principle, there may be a large number of possibilities for creating a vector based on operating parameters. In one embodiment, a first vector is preferably constructed in the same manner as a second vector to which the first vector is to be compared.In the context of the present document, the term "vector-based similarity analysis" can be understood to mean, in particular, a method for calculating the similarity (similarity measure) between vectors in a (multi-dimensional) space (e.g. a coordinate system). The vectors can represent objects (in particular operating parameters), so that ultimately a similarity comparison of these operating parameters can also be carried out, e.g. a comparison of the similarity between an actual vector and a target vector. A large number of methods of vector-based similarity analysis are established (e.g. cosine similarity, Euclidean distance, scalar product), which can be used directly and reliably.In the context of the present document, the term "fluid" is understood to mean, in particular, a liquid and / or a gas, optionally comprising solid particles. The term "fluid" may also refer to a mobile phase in which a fluidic sample is transported. If the viscosity of a fluid is measured, this can be the viscosity of the fluid (mobile phase) itself or the viscosity of the fluidic sample in the mobile phase.In the context of the present document, the term "fluidic sample" is understood to mean, in particular, a medium, further in particular a liquid, which contains the material actually to be analyzed (for example, a biological sample), such as, for example, a protein solution, a pharmaceutical sample, etc.In the context of the present document, the term "mobile phase" is understood to mean, in particular, a fluid, further in particular a liquid, which serves as a carrier medium for transporting the fluidic sample between a fluid drive and a sample separation device. However, a mobile phase can also be used in a fluid conveying device for influencing the fluidic sample. For example, the mobile phase may be a (for example, organic and / or inorganic) solvent or a solvent composition (for example, water and ethanol).In the context of the present document, the term "analysis device" may particularly denote an apparatus capable and configured to examine, in particular separate, further in particular separate into different fractions, a fluidic sample. For example, such sample separation can be carried out by means of chromatography or electrophoresis. Preferably, the analysis device may be a liquid chromatography sample separation apparatus.According to an exemplary embodiment, the invention can be based on the idea that an analysis device can be operated efficiently and reliably if vectors are created based on detected operating parameters and these vectors are then compared with one another by means of vector-based similarity analysis. In this way, anomalies can be found during operation in a particularly efficient and reliable manner. In particular, this approach can significantly increase the operational clarity, both for evaluation software and for a human operator. The vectors created can be shown graphically clearly (e.g. in a coordinate system) and deviations (or the degree of deviation) can be clearly recognized and classified. Preferably, a vector can relate to a time series and / or a plurality of operating parameters, so that a clear (display) result can be generated from a plurality of data. Current data can be easily compared to previous (historical) data to detect an anomaly.Vector similarity analysis of data can improve the reliability and efficiency of analysis devices such as HPLCs by applying similarity calculations to (recurrent) signal data (e.g., pressure signals for each pump stroke, pressure data for each chromatography run using the same method, current signal of the electric motor in the autosampler during each injection process, etc.). Such signal data can be treated as vectors between which different similarity measures can be calculated. In addition, vectors may be constructed from multiple instrument signals at a particular time (e.g., pressure, temperature, and detector values at the start of the run), and the similarity between such vectors may be calculated.The aim here may be to identify unusual or undesirable events and, in particular, to trigger further actions. For example, a non-similar behavior of a current instrument signal under the same operating conditions may result in enhanced data acquisition / analysis or automated reanalysis of the given sample compared to prerecorded instrument signals.With long analysis sequences and routine operations of an analysis device, anomalies are difficult to recognize for the human eye, in particular without viewing all instrument signals visualized in the computer application. Deviations in the operating mode of the device can lead to incorrect results / conclusions, so that problems may remain unrecognized.The invention can aim to detect such anomalies in operation and provide insight into instrument optimization and preventative maintenance. According to the invention, a generic method for detecting exception events during device operation may be employed (in particular without providing further reasoning for the exception event). By aligning operating parameters as vectors and comparing their similarities, it can be made possible to configure the detection of anomalies and unusual events (in recurrent methods) in a simple and efficient manner, but very reliable.EXEMPLARY EMBODIMENTSAccording to an embodiment, the control device is further configured to perform abnormality detection regarding the operation of the analysis device based on the vector-based similarity analysis. This can have the advantage that an anomaly can be detected efficiently and directly, in particular without reference values and / or threshold values (and their configuration) being required.In the present context, the term "anomaly" can mean, in particular, a deviation from the norm or the expected one. An anomaly may be an unusual or unexpected event, behavior, or pattern that is different than the usual events, behavior, or patterns. With regard to data analysis with respect to a system, an anomaly can be, for example, an "outlier" (outlier) and / or a faulty state. In one embodiment, an anomaly may represent a deviation that may be both a relevant (system) fault and harmless observation. In one exemplary embodiment (compare FIG. 2 ), when an anomaly is detected, it may first have to be checked (in particular by a human) whether this is a relevant or irrelevant error.According to one exemplary embodiment, the anomaly has a deviation from a desired operating state, in particular a setpoint operating state, of the analysis device. This can have the advantage that it can be detected quickly and efficiently if there is a deviation from a desired state. Conventionally, a wide variety of data must be reviewed in the operation of analysis devices, so that relevant deviations (with respect to individual parameters) often remain unrecognized (both for humans and for machines). Based on the vector similarity analysis (which can involve a plurality of operating parameters), however, it can often be seen at a glance if there is a deviation from the desired state; speaking graphically, if the arrows of the vectors to be compared deviate significantly from one another.According to one exemplary embodiment, a predefined difference, in particular a threshold value, between the at least two vectors is indicative of the presence of the anomaly. The presence of an anomaly can be determined in various ways. In one example, the vector similarity analysis may be evaluated manually (by a user) and the presence / absence of the anomaly may be decided. In another example, such a selection may be made based on machine learning, in particular artificial intelligence (AI). If the selection is to be made according to defined criteria (in particular automatically), a distinguishing value or value range can be predetermined / defined for this purpose. If the result of the vector similarity analysis is outside (or within, depending on the definition) the predefined difference, this can be indicative of the presence of an anomaly. In a further exemplary embodiment, the predefined difference can also confirm the presence of an anomaly. In a specific embodiment, the predefined difference may be a threshold (threshold).According to one exemplary embodiment, at least one vector is based on a setpoint operating state. This vector can form a reference value with which current vectors are compared. The desired operating state vector can be, for example, a theoretical (ideal) vector and / or a previously measured / detected vector. According to one exemplary embodiment, at least one vector is based on an actual operating state. This current vector can be compared with a desired operating state vector. In this way, an anomaly can be detected or excluded.According to an embodiment, at least one vector is based on a previous operating state ("historical data"). This vector can form a reference value, similar to a desired operating state, with the current vector. In contrast to a theoretical (ideal) target operating state, the previous operating state can relate to a practical (realistic) vector. It can thus be compared whether a stable / desired state changes over time. According to an embodiment, at least one vector is based on a current operating state. This vector can be updated again and again and in each case compared with the previous operating state and / or a desired operating state.According to an embodiment, at least one vector is based on an average operating state. Such an average state may be an indicator of stable (and reliable) operation with which current operating states may be compared. In a simple example, the average operating state may be an arithmetic mean. In a more complex embodiment, other factors such as standard deviations, etc. may be included.According to one exemplary embodiment, the vector-based similarity analysis comprises a comparison of the at least two vectors. This allows differences to be detected efficiently and reliably. In a simple embodiment, the comparison can be carried out visually by an operator, pictorially speaking vectors are viewed and compared in the coordinate system.In a more complex (and / or automated) embodiment, an established mathematical method may be used to provide reliable comparative analysis. This can be, for example, one of the following: angle deviation, cosine similarity, Euclidean distance, scalar product, waterstone distance, Frechet distance, Manhattan distance, Jaccard similarity, Pearson correlation, Hamming distance. These are just a few examples of established methods that can be used directly and reliably.According to an embodiment, the at least one operating parameter comprises at least one of the following (internal factors): pressure (in particular system pressure or pressure at any point in a flow path), temperature, flow rate, detector signal, detector lamp intensity, motor current (e.g. the system pump, the injection pump etc.), pressure per pump cycle, column counterpressure, column temperature, temperature in the sample space (sampler), flow cell temperature (detector), lamp current / voltage (detector), current / voltage of a photosensitive element of a detector (e.g. Photodiode or PMT), encoder signal (motor).According to one exemplary embodiment, the at least one operating parameter comprises at least one of the following (external factors): external factors, for example. Ambient temperature, air humidity (RH), mass / weight sensor signal (e.g. in gravimetric fill level monitoring), pH value of an operating liquid, conductivity of an operating liquid, viscosity of an operating liquid, etc.Accordingly, a plurality of different operating parameters (in particular over time) can be taken into account in order to obtain a specific or comprehensive image of the operation of the analysis device. Multiple operating parameters may each be combined in a vector to increase efficiency and clarity.According to one exemplary embodiment, at least one vector is based on a time profile of the at least one operating parameter. According to an embodiment, at least one vector is created based on two or more operating parameters. According to an embodiment, at least one vector is created based on discrete times and / or as a time series (of one or more operating parameters). This may have the advantage that continuous monitoring (over time) may be provided. This may relate to one or more operating parameters. If a time profile is represented as a vector, a deviation from a setpoint time profile can be detected particularly effectively.According to one exemplary embodiment, at least one vector is created based on at least two operating parameters, in particular considered as a time profile. According to one exemplary embodiment, the at least two operating parameters interact with one another with respect to the operation of the analysis device. In this way, a particular function can be monitored efficiently, for example the actual sample separation in the column furnace. According to one embodiment, these interacting operating parameters relate to at least two of the following: flow rate, column temperature, column back pressure. These operating parameters of the thermostat / column furnace may be of particular relevance to the analysis device and are dependent on each other; thus, may be considered as a whole (e.g., a common vector).According to an embodiment, the detector signal also depends on the temperature (in particular the temperature of the liquid flowing through the flow cell and a stable flow). According to an embodiment, the detector signal also depends on the stability of the lamp, which in turn is also a function of the temperature. In one embodiment, the detector signal may also be integrated into the "column furnace" vector.According to an embodiment, at least one vector is multidimensional. According to an embodiment, at least one vector is based on two or more different operating parameters. This can have the advantage that a plurality of operating parameters (or even whole functions or the entire system) can be represented via a vector. Therefore, many operating aspects can be simultaneously monitored in a particularly efficient manner (less data, less required computer resources).According to one exemplary embodiment, the control device is furthermore configured to generate a data matrix based on the at least one operating parameter, in particular two or more operating parameters. According to one exemplary embodiment, the data matrix has at least one of the at least two vectors. In this way, a plurality of data or operating parameters can be incorporated into the vector similarity analysis.In the present context, the term "matrix" may particularly denote an arrangement of numbers, symbols or expressions in rows and columns. In one example, the mathematical definition of this term may be relevant in the present context. A matrix may be, for example, an m*n matrix with m rows and n columns. A matrix can have a vector, e.g. as an m*1 or n*1 matrix. Accordingly, a matrix can be considered a specific extension of a vector. The vector similarity analysis can thus comprise a matrix similarity analysis in one exemplary embodiment.According to one exemplary embodiment, a point in time for determining the at least one operating parameter comprises at least one of the following: analysis start, analysis end, injection point in time, operation start, operation end, point in time before and / or after a cleaning / rinsing procedure, point in time before and / or after switching a valve, point in time before and / or after carrying out a system maintenance, point in time before and / or after solvent change or refilling of the solvent, point in time when a specific count value is reached (cf. EMF counter number of valve circuits, number of analyses with a specific column, etc.). This can have the advantage that particularly relevant points in time are used for detecting the operating parameters (or time series) in order to contain as much information as possible. For example, the injection time of an analysis device may be particularly suitable because the system is then in equilibrium (equilibrated).According to an embodiment, at least one vector refers to at least one of the following (in particular, monitors at least one of the following): an operating parameter, two or more operating parameters, a periodic profile (anomaly in a periodic profile), the entire system of the analysis device, a method of the analysis device, a functional unit of the analysis device, a module of the analysis device. High flexibility can thereby be provided. Thus, for example, sections of the system (column temperature / pressure) can be monitored or even entire modules (sample separation module) up to the entire system.According to one exemplary embodiment, the analysis device has two or more modules, in particular wherein the analysis device has at least one of the following modules: a pump module, a detector module, a sample separation module, a sample receiving module, a fractionating module. Analysis devices can be organized in a modular manner, in particular an HPLC is usually implemented in the form of modules stacked in the vertical direction. Module-wise monitoring (e.g. one vector per module) may be particularly efficient because an anomaly (or potential fault) may be located quickly.According to an embodiment, at least one vector is superordinate to another vector, in particular concerns a larger / more complex unit. In other words, the vectors can be hierarchically constructed. In this way, larger units can be better reviewed (general review) while anomalies can be more accurately located (detailed specific view). In one embodiment, a first vector may relate to an assembly (e.g., a valve). A superordinate vector can relate, for example, to the module (for example injection module) in that the assembly is integrated. A further superordinate vector can relate, for example, to the entire system (the analysis device).In one embodiment, interleaving of the monitor and ultimately merging into a single value (or reversing breaking down of the sub-monitors for diagnosis) may be implemented. At the lowest level, there may be, for example, vectors for individual signals, for example pressure sensor, furnace temperature, etc. These then flow into superordinate vectors, for example, a module or other functional units intended to monitor specific modules. A system vector could then stand above it and possibly also a fleet / laboratory vector above it. Such vectors across the system could, however, also only depict specific assemblies (e.g. injection valves) in order, for example, to examine the robustness of such an assembly.According to one exemplary embodiment, the control device is further configured to: preprocess data / values with respect to the operating parameter. The pre-processing can comprise e.g. up-sampling, down-sampling, filtering or smoothing. If, for example, a signal with several kHz is recorded, then no vector is required for this over all data points, but already a few Hz may suffice to make a valid statement. In a further exemplary embodiment, interference (e.g. spikes / noise) can be filtered out. This may have the advantage that the data quality is improved and / or the amount of data is reduced.According to an embodiment, the control device is further configured to: trigger an error message (in particular an alarm) when the abnormality is detected. This can quickly react to a potential fault. According to one exemplary embodiment, the control device is further configured to: detect a systematic error during operation of the analysis device. Such an error can often be difficult to detect using conventional methods. With vector-based similarity analysis, such an error can be found more easily and reliably (e.g. by means of a "system" vector).According to an embodiment, the control device is further configured to: mark / flag a vector and / or operating parameter when the anomaly is detected. This can have the advantage that only those data which have potentially been identified as an anomaly have to be analyzed ("review by reception"). This can save time and the operation can initially proceed normally. In one embodiment, this vector may map / monitor a particular system area (e.g., a module or analysis path) and flagging may relate to that portion of the device. Thus, the user can know directly where the fault is located.According to one exemplary embodiment, the control device is furthermore configured to carry out the vector-based similarity analysis and / or the anomaly detection free of at least one of the following: a reference value, a threshold value, a comparison of individual measured values, a consideration of absolute values. This can provide the advantage of efficient and flexible operation. Reference values and / or threshold values must be configured, established and updated as appropriate. This can be cumbersome and less flexible. However, the described approach may operate without specific values by comparing vectors, instead of absolute and / or individual values, that may be indicative of a plurality of operating parameters.According to one exemplary embodiment, the vector-based similarity analysis can be carried out for a period of time X or via X vector analysis data points. This might possibly allow additional findings to be obtained, i.e. the detection of trends / drifts. In one exemplary embodiment, components can be monitored with regard to their wear state or the wear state of the components can be derived. This can bring the advantage that the need for maintenance or replacement with respect to a component can be reliably detected.In the context of the present application, the term "sample separation device" can be understood to mean, in particular, a device for analyzing a fluidic sample, in particular into different fractions. For this purpose, constituents of the fluidic sample can first be adsorbed on the sample separation device and then desorbed separately (in particular fraction by fraction). For example, such a sample separation device can be designed as a chromatographic separation column.According to one embodiment, the analysis device is a sample separation device, in particular a chromatography device, in particular a liquid chromatography device, a gas chromatography device, an SFC (supercritical fluid chromatography) device or an HPLC (high performance liquid chromatography) device.According to one exemplary embodiment, the analysis device is configured as a microfluidic device. According to an embodiment, the analysis device is configured as a nanofluitic device.According to one exemplary embodiment, the sample separation device is designed as a chromatographic separation device, in particular as a chromatographic separation column.According to an embodiment, the fluid actuator is configured to drive the mobile phase and the fluidic sample under high pressure.According to an embodiment, the fluid drive is configured for driving the mobile phase and the fluidic sample with a pressure of at least 500 bar, in particular of at least 1000 bar, further in particular of at least 1200 bar, further in particular of at least 1500 bar.According to one exemplary embodiment, the analysis device has a detector for detecting the analyzed, in particular separated, fluidic sample.According to one embodiment, the analysis device comprises a fractionator for fractionating separate fractions of the fluidic sample.The analysis device may be a microfluidic meter, a life science apparatus, a liquid chromatography apparatus, a gas chromatography apparatus, a high performance liquid chromatography (HPLC), a UHPLC unit or a supercritical fluid chromatography (SFC) apparatus. However, many other applications are possible.According to one exemplary embodiment, the sample separation device can be designed as a chromatographic separation device, in particular as a chromatographic separation column. In chromatographic separation, the chromatographic separation column may be provided with an adsorption medium. At this point, the fluidic sample can be held up and only subsequently can it be separated again in fractions in the presence of a specific solvent composition, whereby the separation of the sample into its fractions is accomplished.A pump system for conveying fluid can be configured, for example, to convey the fluid or the mobile phase through the system at a high pressure, for example a few 100 bar up to 1000 bar and more.The analysis device can have a sample injector for introducing the sample into the fluidic separation path. Such a sample injector can have a sample or injection needle, which can be coupled to a needle seat, in a corresponding liquid path, wherein the sample needle can be moved out of this needle seat in order to receive sample. After reinsertion of the sample needle into the needle seat, the sample may be in a fluid path which may be switched into the separation path of the system, for example, by switching a valve. In another embodiment of the invention, a sample injector or sampler with a sample needle that is operated without a needle seat may be used.The analysis device may comprise a fraction collector for collecting the separated components. Such a fraction collector can lead the various components of the separated sample, for example, into different liquid containers. The analyzed sample can, however, also be supplied to a drain container.Preferably, the analysis device may comprise a detector for detecting the separated components. Such a detector can generate a signal which can be monitored and / or recorded and which is indicative of the presence and amount of the sample components in the fluid flowing through the system.In an exemplary embodiment, the concept of vector similarity analysis is used to determine anomalies in an analytical system (e.g., HPLC). The application of vector similarity analysis for anomaly detection has not been known to date in the field of analytics.In an exemplary embodiment, a function implemented in software or firmware is described for continuous system diagnosis, e.g., for detecting deviations from a current state (anomalies) during operation based on the approach of calculating vector similarity using various mathematical definitions. In comparison to conventional approaches, no reference values or threshold values are provided for the comparison with a current value (in advance, e.g. by the user or read out from a database). Instead, the current(s) value(s) are compared to the last / previous value. However, not the individual values are compared with one another, but the orientation, more precisely the angular deviation, of vectors which represent one or more of the values. Such values may be system parameters such as pressure, temperature, flow rate, detector signal, etc., that may be monitored by sensor signals, the vector representing or consisting of multiple parameters; for example, the state of the entire system or the state of one or more components across multiple systems is expressed by the vector.BRIEF DESCRIPTION OF THE DRAWINGSOther objects and many of the attendant advantages of embodiments of the present invention will be readily appreciated and better understood by reference to the following more detailed description of embodiments taken in conjunction with the accompanying drawings. Features that are substantially or functionally the same or similar are provided with the same reference numerals. FIG. 1 shows an analysis device designed as a sample separation device, according to an exemplary embodiment of the invention. FIG. 2 schematically shows a control of an analysis device according to an exemplary embodiment of the invention.FIGS. 3A to 3C show examples of vector-based similarity analyses based on at least two created vectors, according to exemplary embodiments of the invention.FIGS. 4A to 4C show examples for calculating the similarity analysis, according to exemplary embodiments of the invention.DETAILED DESCRIPTION OF THE DRAWINGSThe illustration in the drawing is schematic.FIG. 1 shows the basic structure of an HPLC system as an example of an analysis apparatus 10 designed as a sample separation device according to an exemplary embodiment of the invention, as can be used for liquid chromatography, for example. A fluid actuator 20 supplied with solvents from a feeder 25 drives a mobile phase through a sample separation device 30 (such as a chromatographic column) containing a stationary phase. The supply device 25 comprises a first fluid component source for providing a first fluid or a first solvent component A (for example water) and a second fluid component source for providing another second fluid or a second solvent component B (for example an organic solvent). An optional degasser 27 may degas the solvents provided by means of the first fluid component source and by means of the second fluid component source before they are supplied to the fluid drive 20. Optionally, the solvents may be mixed at a mixing point.A sample application unit, which can also be referred to as injector 40, is arranged between the fluid drive 20 and the sample separation device 30 in order to initially take a sample liquid or a fluidic sample from a sample container into a sample receiving volume in an injector path, and subsequently introduce it into a fluidic separation path between the fluid drive 20 and the sample separation device 30 by switching an injection valve of the injector 40. The taking up of fluidic sample from the sample container can be effected in particular by a sample needle being moved out of a sample seat and being moved into the sample container, fluidic sample being sucked out of the sample container through the sample needle into the sample receiving volume by means of a fluid conveying device designed as a dosing device, and the sample needle then being moved back into the needle seat.The stationary phase of the sample separation device 30 is provided for separating components of the sample. A detector 50, which may comprise a flow cell, detects separated components of the sample. A fractionating apparatus or fractionator 60 may be provided for dispensing separated components of the sample into dedicated containers. Liquids no longer required can be discharged into a drain container or into a waste line.While a liquid path between the fluid drive 20 and the sample separation device 30 is typically under high pressure, the sample liquid under normal pressure is initially introduced into a region separated from the liquid path, namely the sample loop or the sample receiving volume, of the sample application unit or the injector 40. The sample liquid is then introduced into the separation path under high pressure. A sample loop as sample receiving volume (also referred to as sample loop) can be understood to mean a section of a fluid line which is designed for receiving or buffering a predefined quantity of fluidic sample. Preferably, even before the sample liquid, which is initially under normal pressure, is switched into the separation path which is under high pressure, the content of the sample receiving volume is brought by means of a metering device in the form of the fluid conveying device to the system pressure of the analysis device 10 which is designed as an HPLC.A control device 100 controls the individual components / modules 20, 25, 30, 40, 50, 60, etc. of the analysis device 10. For example, the control device 100 may be coupled to sensors that record operating parameters relating to the modules 20, 25, 30, 40, 50, 60 during operation (e.g., time series of pressure and temperature).The control device 100 is configured to determine operating parameters with respect to the analysis device 10 and to generate at least two vectors with respect to the determined operating parameters. Furthermore, the control device 100 is configured to perform a vector-based similarity analysis based on the at least two generated vectors, whereby, for example, an anomaly during operation can be detected in an efficient and reliable manner.FIG. 2 schematically shows a control of an analysis device 10, e.g. by means of the control device 100 described above, according to an exemplary embodiment of the invention. The automated workflow shown can render continuous human monitoring unnecessary and lead to a more efficient checking method, wherein human intervention is only required when an anomaly is detected. In this embodiment, the operation of the analysis device 10 (via the control device 100) involves a laboratory manager 101 and a laboratory analyzer 102.The laboratory manager 101 configures particular watchdog operating parameters that are to be monitored in particular in order to detect anomalies. The observed signals and the appropriate operating parameters (and optional similarity thresholds) may be configured to correspond to the particular application. The laboratory analyzer 102 delivers the samples to be analyzed during operation of the analysis device 100 and operation starts. As described above, vectors are created on the basis of detected operating parameters and these are then compared with one another by means of vector similarity analysis. According to the watchdog configuration, the performance is monitored and detected parameters are compared with previous data (e.g. actual vector and desired vector) (reference symbol 103).If no anomaly is detected by means of the vector similarity analysis, the operating system (or the control device 100) can operate normally and a report with a minimum amount of data can be output (reference numeral 104).If an anomaly is detected (based on the predefined watchdog configuration), the corresponding data / operating parameters can be marked (or flagged). The flag may trigger reanalysis and / or extended data acquisition (reference 105). The laboratory manager will then analyze the flagged data / operating parameters and make a decision (possibly using the additional data) as to whether the system is functioning as desired or whether an active problem must be solved (reference numeral 106).In other words, as soon as an anomaly is detected, it is identified and a "review by reception" process is triggered. This may include extended data collection and / or automated suspect reanalysis for extended data analysis and / or manual verification of flagged data to remedy problems.Similarities and anomalies can be difficult to estimate with the human eye, especially when the data becomes multi-dimensional. The described approach provides a generic way to identify such anomalies by comparing similarity to one / more / all other signal events under the same or similar operating methods without having to define thresholds or reference signals.FIGS. 3A to 3C show examples of vector-based similarity analyses based on at least two created vectors 110, 120, according to exemplary embodiments of the invention. The alignment of signals as vectors and the comparison of their similarities in recurrent methods can be applied to several parts of the analysis apparatus and its processes. Based on processes observed in the form of instrument signals and parameterized by device settings, vectors can be constructed that are compared using similarity measures. Three illustrative examples are given below:FIG. 3A : angle θ near 0°, cos(θ) near 1, similar vectors 110, 120.FIG. 3B : angle θ close to 90°, Cos(θ) close to 0, orthogonal vectors 110, 120.FIG. 3C: angle θ close to 180°, Cos(θ) close to -1, opposing very dissimilar vectors 110, 120.FIGS. 4A to 4C show examples for calculating the similarity analysis, according to exemplary embodiments of the invention. For similarity-based data analysis, the vector of one instance is compared to one / more / all other vectors of other instances of this process. FIG. 4A shows the example of the similarity calculation by means of cosine similarity, FIG. 4B by means of Euclidean distance and FIG. 4C by means of the dot product (dot product). This information may be useful when comparing recurrent procedures in an analytical device such as an HPLC instrument at multiple different stages to determine exceptions / anomalies.Reference numerals denote reference numerals10 Analysis device 20 Fluid drive 25 Feeder 27 Degasser 30 Sample separator 40 Injector 50 Detector 60 Fractionator 100 Control device 101 Laboratory manager 102 Laboratory analyzer 103 "Watchdog" configuration 104 No abnormality detected 105 Abnormality detected 106 Manual analysis of abnormality 110 First vector 120 Second vectorReferences included in the specificationThis list of documents cited by the applicant has been produced in an automated manner and is only included for the better information of the reader. The list is not part of the German patent application or utility model application. The DPMA does not take any adhesion for any faults or omissions.Patent Literature citedEP 0,309,596 B1

[0002]

Claims

An analysis device (10), in particular a sample separation device, for carrying out an analysis, which has a control device (100) which is configured to: determine at least one operating parameter with respect to the analysis device (10); generate at least two vectors (110, 120) with respect to the at least one operating parameter; and carry out a vector-based similarity analysis based on the at least two generated vectors (110, 120).The analysis device (10) according to claim 1, wherein the control device (100) is further configured to: perform abnormality detection regarding the operation of the analysis device (10) based on the vector-based similarity analysis.The analysis device (10) according to claim 1 or 2, wherein the anomaly comprises a deviation from a desired operating state, in particular a target operating state, of the analysis device (10).The analysis device (10) according to one of the preceding claims, wherein a predefined difference, in particular a threshold value, between the at least two vectors (110, 120) is indicative of the presence of the anomaly.The analysis device (10) according to one of the preceding claims, wherein the at least two vectors (110, 120) have at least one of the following features: at least one vector is based on a desired operating state; at least one vector is based on an actual operating state; at least one vector is based on a previous operating state; at least one vector is based on a current operating state; at least one vector is based on an average operating state.The analysis device (10) according to one of the preceding claims, wherein the vector-based similarity analysis comprises a comparison of the at least two vectors (110, 120), in particular by means of at least one of the following: angle deviation, cosine similarity, Euclidean distance, scalar product, waterstone distance, Frechet distance, Manhattan distance, Jaccard similarity, Pearson correlation, Hamming distance.The analysis device (10) according to any of the preceding claims, wherein the at least one operating parameter comprises at least one of the following: pressure, temperature, flow rate, detector signal, detector lamp intensity, motor current, pressure per pump cycle, column back pressure, column temperature, temperature in the sample space, flow cell temperature, lamp current / voltage.The analysis device (10) according to one of the preceding claims, wherein at least one vector (110, 120) is created based on a time profile of the at least one operating parameter, in particular two or more operating parameters, in particular with respect to discrete points in time and / or as a time series.The analysis device (10) according to one of the preceding claims, wherein at least one vector (110, 120) is created based on at least two operating parameters, in particular viewed as a time profile, in particular wherein the at least two operating parameters interact with one another with respect to the operation of the analysis device (10), further in particular at least two of the following: flow rate, column temperature, column back pressure, detector signal.The analysis device (10) according to any one of the preceding claims, wherein at least one vector (110, 120) is multidimensional, in particular based on two or more different operating parameters.The analysis device (10) according to one of the preceding claims, wherein the control device (100) is further configured to: create a data matrix based on the at least one operating parameter, in particular two or more operating parameters, wherein the data matrix comprises at least one of the at least two vectors (110, 120).The analysis device (10) according to any one of the preceding claims, wherein a time for determining the at least one operating parameter comprises at least one of the following: analysis start, analysis end, injection time, operation start, operation end, time before / after a cleaning / rinsing procedure, time before / after switching a valve, time before / after performing a system maintenance, time before / after a solvent change.The analysis device (10) according to any one of the preceding claims, wherein at least one vector (110, 120) relates to at least one of the following, in particular monitors at least one of the following: an operating parameter, two or more operating parameters, a periodic profile, the entire system of the analysis device (10), a method of the analysis device (10), a functional unit of the analysis device (10), a module (20, 30, 40, 50, 60) of the analysis device (10), in particular wherein the analysis device (10) has two or more modules, further in particular wherein the analysis device (10) has at least one of the following modules: a pump module (20), a detector module (50), a sample separation module (30), a sample receiving module (40), a fractionating module (60).The analysis device (10) according to any one of the preceding claims, wherein the control device (100) is further configured to: preprocess data on which the at least one operating parameter is based, in particular by means of at least one of up-sampling, down-sampling, filtering, smoothing.The analysis device (10) according to any one of the preceding claims, wherein the control device (100) is further configured to: trigger an error message when the anomaly is detected; and / or flag a vector (110, 120) and / or operating parameter when the anomaly is detected; and / or detect a systematic error in the operation of the analysis device (10).The analysis device (10) according to any one of the preceding claims, wherein the control device (100) is further configured to perform the vector-based similarity analysis and / or the anomaly detection free of at least one of the following: a reference value, a threshold value, a comparison of individual measurement values, a consideration of absolute values.The analysis device (10) according to any one of the preceding claims, further comprising at least one of the following features: the analysis device (10) comprises a fluid drive (20) for driving a mobile phase and a fluidic sample injected into the mobile phase; the analysis device (10) comprises a sample separation device (30) for separating the fluidic sample injected into the mobile phase; the analysis device (10) is configured for analyzing at least one physical, chemical and / or biological parameter of the fluidic sample; the analysis device (10) is configured as a sample separation device for separating the fluidic sample; the analysis device (10) is a chromatography device, in particular a liquid chromatography device, a gas chromatography device, an SFC (supercritical fluid chromatography) device or an HPLC (high performance liquid chromatography) device; the analysis device (10) is configured as a microfluidic device; the analysis device (10) is configured as a nanofluous device; the sample separation device (30) is configured as a chromatographic separation device, in particular as a chromatographic separation column; the fluid drive (20) is configured for driving the mobile phase and the fluidic sample under high pressure; the fluid drive (20) is configured for driving the mobile phase and the fluidic sample at a pressure of at least 500 bar, in particular of at least 1000 bar, further in particular of at least 1200 bar; the analysis device (10) has a detector (50) for detecting the analyzed, in particular separated, fluidic sample; the analysis device (10) has a fractionator (60) for fractionating separated fractions of the fluidic sample.A computer-implemented method for operating an analysis device (10), in particular a sample separation device, the method comprising: determining at least one operating parameter with respect to the analysis device (10); creating at least two vectors (110, 120) with respect to the at least one operating parameter; and carrying out a vector-based similarity analysis based on the at least two created vectors (110, 120).A data processing apparatus configured to carry out the method of claim 18.using vector-based similarity analysis to detect an abnormality regarding the operation of an analysis device ( 10).

Images