CN114935548B - One-time operation detection method for detecting multiple types of indexes of oil sample - Google Patents

One-time operation detection method for detecting multiple types of indexes of oil sample Download PDF

Info

Publication number
CN114935548B
CN114935548B CN202210852758.2A CN202210852758A CN114935548B CN 114935548 B CN114935548 B CN 114935548B CN 202210852758 A CN202210852758 A CN 202210852758A CN 114935548 B CN114935548 B CN 114935548B
Authority
CN
China
Prior art keywords
detection
oil
oil sample
model
reflectivity
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202210852758.2A
Other languages
Chinese (zh)
Other versions
CN114935548A (en
Inventor
方彦
罗元辉
高韩
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Lianqiao Network Cloud Information Technology Changsha Co ltd
Original Assignee
Lianqiao Network Cloud Information Technology Changsha Co ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Lianqiao Network Cloud Information Technology Changsha Co ltd filed Critical Lianqiao Network Cloud Information Technology Changsha Co ltd
Priority to CN202210852758.2A priority Critical patent/CN114935548B/en
Publication of CN114935548A publication Critical patent/CN114935548A/en
Application granted granted Critical
Publication of CN114935548B publication Critical patent/CN114935548B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
    • G01N21/17—Systems in which incident light is modified in accordance with the properties of the material investigated
    • G01N21/25—Colour; Spectral properties, i.e. comparison of effect of material on the light at two or more different wavelengths or wavelength bands
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N11/00—Investigating flow properties of materials, e.g. viscosity, plasticity; Analysing materials by determining flow properties
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N15/00—Investigating characteristics of particles; Investigating permeability, pore-volume or surface-area of porous materials
    • G01N15/02—Investigating particle size or size distribution
    • G01N15/0205—Investigating particle size or size distribution by optical means
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N33/00—Investigating or analysing materials by specific methods not covered by groups G01N1/00 - G01N31/00
    • G01N33/26—Oils; Viscous liquids; Paints; Inks
    • G01N33/28—Oils, i.e. hydrocarbon liquids
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N33/00—Investigating or analysing materials by specific methods not covered by groups G01N1/00 - G01N31/00
    • G01N33/26—Oils; Viscous liquids; Paints; Inks
    • G01N33/28—Oils, i.e. hydrocarbon liquids
    • G01N33/2835—Specific substances contained in the oils or fuels
    • G01N33/2858—Metal particles
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N5/00—Analysing materials by weighing, e.g. weighing small particles separated from a gas or liquid
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N9/00—Investigating density or specific gravity of materials; Analysing materials by determining density or specific gravity
    • G01N9/02—Investigating density or specific gravity of materials; Analysing materials by determining density or specific gravity by measuring weight of a known volume
    • G01N9/04—Investigating density or specific gravity of materials; Analysing materials by determining density or specific gravity by measuring weight of a known volume of fluids
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N9/00—Investigating density or specific gravity of materials; Analysing materials by determining density or specific gravity
    • G01N9/36—Analysing materials by measuring the density or specific gravity, e.g. determining quantity of moisture
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G06F17/18—Complex mathematical operations for evaluating statistical data, e.g. average values, frequency distributions, probability functions, regression analysis
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N11/00—Investigating flow properties of materials, e.g. viscosity, plasticity; Analysing materials by determining flow properties
    • G01N2011/006—Determining flow properties indirectly by measuring other parameters of the system
    • G01N2011/008—Determining flow properties indirectly by measuring other parameters of the system optical properties

Landscapes

  • Chemical & Material Sciences (AREA)
  • Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • General Physics & Mathematics (AREA)
  • Health & Medical Sciences (AREA)
  • Engineering & Computer Science (AREA)
  • Analytical Chemistry (AREA)
  • Pathology (AREA)
  • Immunology (AREA)
  • General Health & Medical Sciences (AREA)
  • Biochemistry (AREA)
  • Data Mining & Analysis (AREA)
  • Pure & Applied Mathematics (AREA)
  • Oil, Petroleum & Natural Gas (AREA)
  • General Chemical & Material Sciences (AREA)
  • Medicinal Chemistry (AREA)
  • Food Science & Technology (AREA)
  • Chemical Kinetics & Catalysis (AREA)
  • Theoretical Computer Science (AREA)
  • Computational Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Mathematical Analysis (AREA)
  • Mathematical Optimization (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Probability & Statistics with Applications (AREA)
  • Operations Research (AREA)
  • Algebra (AREA)
  • Evolutionary Biology (AREA)
  • Databases & Information Systems (AREA)
  • Software Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Spectroscopy & Molecular Physics (AREA)
  • Dispersion Chemistry (AREA)
  • Investigating Or Analysing Materials By Optical Means (AREA)

Abstract

The invention discloses a one-time operation detection method for detecting various indexes of an oil sample, which comprises the steps of establishing a relation between an oil liquid collecting point and a detection index; the algorithm model server establishes an algorithm model base according to oil performance, brand, serial number, operating equipment, detection components and acquisition points; implanting the oil to be detected into a cuvette of a detection instrument from a sampling container, giving a serial number to the oil sample, and generating the reflectivity and amplitude brightness value of the oil sample by using an optical system of the hyperspectral oil detection equipment; binding the reflectivity and the amplitude brightness value with a collection point, and uploading the reflectivity and the amplitude brightness value together with the oil sample number and the detection time to an algorithm model server; the application software selects different model algorithms according to the acquisition points; and (6) summarizing detection data. The method can simultaneously obtain the detection results of the metal components, the granularity, the viscosity and the chemical components by one-time operation, achieves the effects of simplifying the operation, saving the consumables, carrying out the real-time detection and saving the links of the operation of professionals.

Description

One-time operation detection method for detecting multiple types of indexes of oil sample
Technical Field
The invention belongs to the technical field of spectral analysis, and particularly relates to a one-time operation detection method for detecting various indexes of an oil sample.
Background
The method for quantitatively detecting the amount of metal-containing dust and chemical components in the oil by adopting a hyperspectral reflection technology of a spectrum band of 400-1000 nm breaks through the traditional emission spectrum analysis method (atomic spectrometer technology), simplifies the detection equipment and becomes portable (1 kg). The detection speed is high, and all detection results can be obtained within 5-6 seconds. Simple and convenient operation and no need of professional personnel. Low cost operation and maintenance (low material consumption), and no material consumption when the computer is started up and corrected every day. However, the adoption of the hyperspectral reflection technology to detect the amount of metal-containing dust and chemical components in oil is a huge challenge, is limited by the range of the collected reflection spectrum wave band, and is insufficient in the information content of the metal and the chemical components through the existing spectrum wave band identification and quantitative detection. Although a group of spectrum wave bands corresponding to each detected element can be known, the spectrum wave bands are beneficial to being split from the collected hyperspectrum, modeling can be carried out according to the detected oil sample and the content distribution of the detected components, the characteristic value of the detected oil sample is calculated, a statistic and deduction regression algorithm is provided, and therefore the detection index result is obtained through rapid convergence. The laboratory for detecting the metal components, granularity, viscosity and chemical components (phosphorus and boron) of the collected oil sample is completed by 4 different devices and operations. For example, on a ship, a laboratory cabin needs to be specially arranged, and a professional needs to be configured to carry out the operation on the ship. The collected oil sample needs to be detected in various aspects including metal components (equipment abrasion position), granularity (abrasion degree), viscosity (additive quality), chemical components (engine water inflow) and the like. The method for simultaneously obtaining the detection results of the metal components, the granularity, the viscosity and the chemical components is explained through the spectral characteristics of the detected oil sample, the information of the oil sample collection point, machine learning and one-time operation. The effects of simplifying operation, saving material and carrying out real-time detection are achieved. The laboratory was moved to the site. The link of operation of professional personnel is saved.
The hyperspectral oil detection principle adopts a 400nm-1000nm halogen light source, a detected oil sample is filled in a rectangle with the volume of 3.5ml, a cuvette with two light transmission surfaces (the applicable wavelength is 350nm-2000nm, and the light transmission ratio is plus 90 percent) is inserted into a darkroom of hyperspectral oil detection equipment, and a spectrum with a specific wavelength, also called hyperspectral characteristic spectral line, is obtained through a reflection light path. The reflected light is generally represented by a characteristic line of the ion cluster (molecule). The Full width at half maximum FWHM of the peak of the spectrum section is wide, element (atom) spectral lines are narrow and approximate to lines, and the spectral characteristics of viscosity, granularity, chemical components and the like are hidden in the spectrum section. Because the characteristic spectral lines of two different elements do not completely coincide in the nature, the element spectral lines are extracted from the natural environment to realize the identification and quantitative analysis of the detected elements. When more than one detected element exists in the detected oil sample, a series of spectral lines with various wavelengths corresponding to each element appear in the spectrum, and the spectral lines generally have a range of dozens of spectral lines. And these spectral lines are mixed and even overlapped with spectral lines of other elements. These spectral lines must be separated to extract the spectral lines of the target element to enable identification and quantitative analysis of the element. In the process of spectral analysis, a plurality of spectral lines with main characteristics in the spectral line range are usually selected for model calculation, so that the detection of target element components and content is realized. Even so, hyperspectral oil detection consists of a set of optical systems and algorithmic models. The accuracy and the reliability of an optical system are improved, and an algorithm model enables the algorithm to quickly and linearly regress (converge) according to the acquisition of main characteristic spectral lines such as elements, ion clusters, particle sizes or viscosity, and the quantitative detection capability of hyperspectral oil detection is determined by solving multiple correlations and consistency.
Disclosure of Invention
In view of the above, the invention provides a method for simultaneously obtaining the detection results of metal components, granularity, viscosity and chemical components through spectral characteristics of the detected oil sample, acquisition point information of the oil sample, machine learning and one-time operation.
The invention discloses a one-time operation detection method for detecting various indexes of an oil sample, which comprises the following steps:
establishing a relation between an oil liquid collecting point and a detection index, wherein the collecting point is that the content of collected oil sample components changes along with different collecting time, and the detection index is an algorithm model corresponding to hyperspectral oil liquid detection equipment;
the algorithm model server establishes an algorithm model library according to oil performance, brand, serial number, operating equipment, detection components and acquisition points;
the detected oil is implanted into a cuvette of a detection instrument from a sampling container, an oil sample number UID is given, and the reflectivity and amplitude brightness value of the oil sample are generated by an optical system of the hyperspectral oil detection equipment. Binding the reflectivity and the amplitude brightness value with an acquisition point, and uploading the reflectivity and the amplitude brightness value together with the oil sample number and the detection time to an algorithm model server;
the application software selects different model algorithms according to the acquisition points: when a collection point is selected in the detection operation, the application software determines the reflectivity and amplitude brightness value obtained in the collection operation according to the setting and calls one or more model algorithms, multi-threads are started according to different model algorithms, and meanwhile, the reflectivity and the amplitude brightness value of the detected oil sample are pushed to each thread to carry out model calculation;
summarizing detection data, monitoring the running of all threads, ensuring that the operation of the last algorithm model thread is finished, and generating a result; and then summarizing the operation results of all threads, and pushing the results to a terminal according to data and report formats.
Further, when the application software calls the model for detection at the 1 st time, the model is used for carrying out self-adaptive learning in advance according to the oil sample bottom; the self-adaptive learning is carried out by adopting a folding subset staggered prediction response method and partial least square method modeling prediction, namely, the partial least square method modeling prediction is adopted, part of subsets are used as observation, the number of the subsets is related to the distribution gradient of the modeling oil sample components, and the result response is measured by mean square error.
Further, dividing the collected data into a training set and a test set according to the modeling dilution gradient distribution or dependent variable;
only using data in the training set to train and perfect the model, then using the model to predict the test set, and calculating a response test MSE;
repeating the step K times, using different training sets and test sets each time, and enabling the predicted value of the model to be closer to the output of the model training set according to the iterative training times;
the average of the K test MSEs is taken as the overall test Mean Square Error.
Further, diluting the actual collection point oil sample with the laboratory detection result according to the concentration gradient by combining the bottom oil through a volume-specific gravity method to obtain a group of known distribution oil sample groups;
using hyperspectral oil detection equipment to generate a set of amplitude brightness values and reflectivity of the oil sample group, wherein each set of amplitude brightness values corresponds to a set of reflectivity according to a spectrum section,
will divide test set K i The reflectivity and amplitude brightness series are input into the model one by one. The model analyzes the statistical relationship between a dependent variable and an independent variable by adopting a partial least square method, wherein the dependent variable Y is the metal component of the oil sample at a certain dilution distribution point at the collection point;
calculating the detection index and the index concentration of the dilution distribution point for calibration, and iteratively converging the detection index and the index concentration of the dilution distribution point to a Mean Square Error specified range;
repeating the steps for i times, wherein 1< i < K, K starts from 0, different training sets and test sets are used each time to represent different gradients of oil sample dilution distribution, and the predicted value of the model is closer to the output of the model training set according to the iterative training times.
Further, the partial least square method comprises the following specific steps:
building a residual information matrixE 0 And a matrix of detected oil sample componentsF 0 In whichE 0 For the normalized independent variable matrix, each row is a series of component indexes, and each column represents a group of spectrum variables corresponding to the detection element indexes;F 0 is a dependent variable matrix; also inE 0 Each row is a series of component indices, and each column represents a set of spectral range variables corresponding to a test element index; data normalization, i.e. subtracting the mean value of each spectrum, and then dividing by the standard deviation of each spectrum;
solution matrix
Figure 148313DEST_PATH_IMAGE001
Feature vector corresponding to the maximum feature value ofw 1 Calculating a component score vector
Figure 771228DEST_PATH_IMAGE002
And a residual information matrix
Figure 471331DEST_PATH_IMAGE003
Wherein
Figure 839864DEST_PATH_IMAGE004
；
Solution matrix
Figure 232799DEST_PATH_IMAGE005
Feature vector corresponding to the maximum feature value of (2)w 2 Calculating a component score vector
Figure 319573DEST_PATH_IMAGE006
And a residual information matrix
Figure 88946DEST_PATH_IMAGE007
In which
Figure 62718DEST_PATH_IMAGE008
；
Repeating the above steps to the firstmStep, solution matrix
Figure 875822DEST_PATH_IMAGE009
Feature vector corresponding to the maximum feature value ofw m Calculating a component score vector
Figure 200624DEST_PATH_IMAGE010
；
Determining co-extraction based on cross-validationmAn ingredientt 1 , t 2 , …, t m Obtaining a satisfactory prediction model, and solvingF 0 In thatt 1 ,t 2 ,…,t m General least squares regression equation above:
Figure 491797DEST_PATH_IMAGE011
wherein the content of the first and second substances,
Figure 585655DEST_PATH_IMAGE012
the weighting parameters of the 1 st, 2 nd and mth components respectively,Fmthe residual information matrix after the m components are extracted;
if data tables X and Y are finally extracted from XmA component of
Figure 569661DEST_PATH_IMAGE013
，
Figure 116180DEST_PATH_IMAGE014
Substitution into
Figure 476623DEST_PATH_IMAGE015
，Namely obtainpPartial least squares regression equation for each dependent variable:
Figure 424987DEST_PATH_IMAGE016
here, the
Figure 579894DEST_PATH_IMAGE017
Satisfy the requirement of
Figure 879288DEST_PATH_IMAGE018
，
Figure 246684DEST_PATH_IMAGE019
，
IIs a dependent variablejIn response to the detected index tag parameter,hthe number of dimensions Y, i.e. the number of spectral segments,
Figure 580714DEST_PATH_IMAGE020
is a model matrix parameter, whereinjRepresents the index of the components,nrepresents the index of the spectral band,
Figure 906522DEST_PATH_IMAGE021
is as followskRelative index of detected componentnResidual information feature vectors for spectral fragments.
Further, the bulk specific gravity method comprises the following steps:
calculating the specific gravity of the oil sample by injecting 10ml of oil sample into a test tube through weight difference to obtain the specific gravity of the bottom oil and the oil sample with laboratory detection results;
obtaining the weight of the cuvette needing to dilute the background oil and detect the oil sample: calculating the weight of 2 oil samples which need to be injected into a 3.4ml cuvette and mixed according to the weights of two 10ml different oil samples;
respectively injecting the weights of the basic bottom oil and the detected oil sample into the cuvette according to the calculation requirements of the dilution point, wherein the calculation method comprises the following steps:
Figure 427633DEST_PATH_IMAGE022
Figure 864299DEST_PATH_IMAGE023
the weight of the detected oil sample is injected according to the target dilution concentration,
Figure 787256DEST_PATH_IMAGE024
the weight of an oil sample to be injected into the cuvette is used, the target dilution concentration is a dilution point of the oil sample detected relative to the detection result of the attached laboratory, and the detection component is a target component in the oil sample detected according to the detection result of the attached laboratory;
the base oil needs to be added into the cuvette by the following weight:
Figure 283965DEST_PATH_IMAGE025
；
placing the cuvette into a support and placing the cuvette on an electronic balance to obtain static weight;
and (4) injecting the oil sample and the background oil respectively by using a liquid transfer device according to the calculated weights of the detection oil sample and the background oil to obtain the detection oil sample of the reconstructed dilution point.
The invention has the following beneficial effects:
the method can simultaneously obtain the detection results of metal components, granularity, viscosity and chemical components by one-time operation, achieves the effects of simplifying operation, saving consumables, carrying out portable real-time detection and saving the links of operation of professionals.
Drawings
FIG. 1 is a comprehensive detection flow chart of hyperspectral oil detection technical equipment;
FIG. 2 is a diagram of data split into subsets;
FIG. 3 is a schematic diagram of the separation of collected data into a training set and a prediction set;
FIG. 4 is a schematic diagram of iterative refinement of a prediction model and observation feedback.
Detailed Description
The invention is further described with reference to the accompanying drawings, but the invention is not limited in any way, and any alterations or substitutions based on the teaching of the invention are within the scope of the invention.
The hyperspectral oil detection technology comprises three core parts. 1) A light source. Halogen light is projected into a detected oil sample in a darkroom cuvette container, and ion clusters (molecular particles) in the oil sample reflect characteristic spectra. 2) An optical system. And collecting, distinguishing and identifying characteristic spectral lines corresponding to specific ion clusters in the reflection spectrum, and dividing the characteristic spectral lines into wave bands with the precision of 2nm by using the optical splitter. 3) And combining a spectral model built by an application scene and a data processing algorithm. And establishing a spectrum model according to the using period of the oil and fat of the oil to be detected by knowing the corresponding relation between the spectrum wave band and the detection element, analyzing and processing the characteristic spectrum line of the oil sample to be detected by an algorithm, quantitatively calculating a detection result, and displaying according to an actual application scene (an expert system) to provide a diagnosis result.
The hyperspectral oil detection technology adopts a reflection light path system, obtains the reflection spectrum of the detected oil ion cluster by the projection of halogen light at a fixed angle, extracts atomic spectral lines (one group) by a mathematical method, and simplifies the structural design, process and material consumption of an atomic excitation light source. The requirements on the range of a spectrum wave band are reduced through modeling and deducing a measurement algorithm for the characteristics and the range of the detected oil, the dependence on the ultra-violet spectrum range is avoided, and the complexity of a spectrum system is greatly reduced. The system is portable, low in cost, real-time and intelligent.
According to the output of the collection and optical system, the reflectivity and DN value (amplitude brightness value) of the detected oil sample according to the spectrum segment are obtained. Assuming that the relationship between the emission spectrum of the detected element and the element concentration corresponding to the intensity of each spectral line is known, the concentration level (PPM) numerical label of the detected element in the detected oil sample can be calculated. If it is stored in the system databaseThere are a sufficient number of oil sample component density distributions (all detected component density) of different metal element content concentrations and their corresponding spectra, such as models of several oil sample samples operating between 100 hours and 200 hours at start-up. The detected oil samples between them can be calculated by Partial Least Square principle (Partial Least Square) with spectral band n as independent variable
Figure 557952DEST_PATH_IMAGE026
The detection index p is a dependent variable
Figure 798309DEST_PATH_IMAGE027
And (4) relationship. The data table of the independent variable and the dependent variable is formed by observing the detected oil sample parameters (the model of the oil sample) among a plurality of known oil sample points in the system database according to the statistical relationship between the dependent variable and the independent variable
Figure 575772DEST_PATH_IMAGE028
And
Figure 243383DEST_PATH_IMAGE029
. Partial least squares regression extracts the 1 st component t from X and Y according to the index of the oil sample to be detected and the corresponding spectral band, based on the analysis ability of the independent variable component to the dependent variable component (the detection index corresponds to the known spectral band) 1 And u 1 . Partial least squares regression separately performed X for t 1 Regression of and Y for u 1 And (4) regression of (1). If the regression equation achieves satisfactory results (accuracy or trend of change is maximized), the algorithm terminates. If satisfactory results are not achieved, X will be used and t will be used 1 The interpreted residual information and Y is u 1 The interpreted residual information is subjected to the extraction of the 2 nd round of components. The iteration is repeated until a satisfactory accuracy is achieved. Such as a 160 hour run of the oil sample under test, the spectra (reflectance and amplitude values) obtained. If m components are finally extracted from the spectrum X
Figure 473507DEST_PATH_IMAGE030
Wave band, offsetLeast squares regression will be performed
Figure 517555DEST_PATH_IMAGE031
(an index of a certain element component) pair
Figure 149525DEST_PATH_IMAGE030
And (4) carrying out inversion calculation on the regression of the wave band to obtain a certain element index of the detected oil sample.
And (3) inputting the spectrum (the reflectivity and the amplitude brightness value after splitting) of the detected oil sample by a data processing and quantitative calculation algorithm according to the parameters of the spectrum model, and quantitatively calculating a detection result by regression inversion of a partial least square method. In practice, the aim of detection of an application scene is combined, an oil sample is actually collected, and a corresponding detection result of a laboratory is used for carrying out one-time calibration on the model. According to the method, the reflectivity and the amplitude brightness value of the spectrum section of the oil sample detected at one time are subjected to spectrum section characteristic splitting, calculation and analysis respectively aiming at different models through different model combinations, so that corresponding detection results are obtained. Such as metal content, particle size, viscosity and chemical content. Thereby reducing detection operation and speeding up detection.
The spectral model is built by a group of oil sample samples reflecting the oil change of the actual application scene corresponding to the spectral characteristics of the oil sample samples. The oil sample covers the whole life cycle of the oil in the equipment, such as the whole cycle from oil filling to oil changing of the lubricating oil in the rotating equipment. And (4) generally selecting 20 to 30 oil sample samples for modeling according to a specific application scene. In practical application, the test accuracy is achieved by obtaining high-quality oil sample samples or by diluting and mixing standard oil and actual oil samples for modeling. The patent mainly describes how to apply an algorithm model established according to different detection index classes, and machine learning is adaptive to the oil sample background. The cross-index type detection effect is achieved through one detection operation. The modeling principles and methods are referenced in the relevant patent literature.
Proper operation of heavy asset equipment is critical to the owning manufacturer. The health state of the equipment is evaluated, managed and maintained through active operation and maintenance, so that the equipment can keep normal operation. Oil-liquid analysis in mechanical equipment is a key link and technology for detecting the health state of the equipment. The oil detection supplements vibration analysis, thermal imaging and other predictive maintenance technologies to monitor, diagnose and evaluate the health state of the equipment. However, oil detection is a complex physical and chemical process, and most of the oil detection is still based on field collection and is sent to a laboratory for detection. The mechanical conditions can vary significantly in the time required for the laboratory to return the oil sample results. Such as an aircraft engine. In certain circumstances, the detection criteria relate to metal composition, particle size, viscosity and chemical composition. Such as lubricating oil, hydraulic oil, etc. The hyperspectral oil detection technical equipment has the advantages of on-site oil analysis and real-time detection, eliminates the waiting, realizes comprehensive detection, can assist in instant decision-making, and has very important significance.
The hyperspectral oil detection technical principle is based on the reflectivity and amplitude brightness value obtained by a photoelectric sensor from a detected oil sample, and the detected oil component is deduced through a model algorithm. The model is formed by modeling the detection component result of a known oil laboratory based on the detected oil sample components. The same spectrum (400 nm-1000 nm) spectrum band, reflectivity and amplitude brightness value are used for obtaining corresponding detection results from the same detected oil sample through different model corresponding algorithms. For one detection operation, how to integrate the above mechanisms, different detections are completed, and the result shows that the detection can be completed through the following operation flow.
1) And establishing a relation between the acquisition points and the detection indexes. The pick-up point is fixed relative to the apparatus and the oil used therein. Only the acquisition time was varied. The collected oil sample has different oil sample component content changes along with the collection time. The detection index refers to an algorithm model corresponding to hyperspectral oil detection equipment. The algorithm model is different according to operation equipment (acquisition points), oil brand, serial number and detection index types (such as metal components, granularity, viscosity and chemical components). The detection index and type are fixed relative to the acquisition point. A corresponding relation is established between operating equipment, collecting points, detected oil and detected indexes through hyperspectral oil detection equipment system software (application software).
2) And binding the detected oil sample with a collection point. The collection point and the collection oil sample container (e.g., cuvette) are bound at the time of the collection operation. The collection points are fixed, and the collected oil sample containers are random, so that the relation can be established only by the number and the name of the collection container or the UID of the container and the collection points during collection. The collecting point is fixed relative to the equipment and the detected oil, and can be naturally bound with the oil algorithm model. The operation is based on establishing a relation between the collection point and the detected oil sample model through application software during initialization. It is understood that one acquisition point may correspond to multiple model algorithms. Meaning that the collection points need to be different types of components to be detected. Such as both metallic and chemical components.
The application software runs in a cloud server or on an off-line computer, and performs distinguishing management, equipment management (hyperspectral oil detection equipment), data management (report generation, trend change, traceability) and the like on the brand, the manufacturer, the type and the like of the detected oil sample.
3) And (4) implanting the detected oil liquid into a cuvette from the sampling container, and giving an oil sample number (UID). The cuvette filled with the oil sample is inserted into a cuvette darkroom of hyperspectral oil detection equipment, a collection point is selected through a man-machine interaction page, a collected oil sample number (option) is input, and detection operation is executed. And the hyperspectral oil detection equipment optical system generates the reflectivity and amplitude brightness value of the oil sample. And binding the reflectivity and the amplitude brightness value with an acquisition point and uploading the oil sample number (or detection time) to an algorithm model server. After a few seconds, the detection result required by the acquisition point is returned and displayed. The detection result can be retrieved, traced to the source and sorted by the number (or the detection time). And the oil sample number is bound with the acquisition points and the acquisition time, and the detection results of all acquired data of the acquisition points can be sorted and displayed according to the time. The change trend of the oil at the collection point is clear at a glance. How the system pushes the oil samples to different model algorithms depending on the acquisition point is described in detail below.
4) The model algorithm is different according to oil performance, brand, row number, operation equipment and detection components. According to application scenarios, collection points are sometimes sensitive, and a special algorithm is required to 'correct' an existing model or train the model through machine learning so as to adapt to the oil sample background (performance, viscosity, pollution degree, brand and ranking). This method is discussed in detail in the next section. It is to be understood here that the model is a parameter matrix a,
Figure 988037DEST_PATH_IMAGE032
（1）
wherein n represents a spectral reflection band of 400nm-1000nm and has 300 discrete spectral lines. m represents the number of indices for detecting oil sample constituents, such as 24 metal constituents. The model of the algorithm is used as a model,
Figure 971036DEST_PATH_IMAGE033
wherein Y is a mixed reflection spectrum vector, X is a detection index vector, A is a model matrix,
Figure 553196DEST_PATH_IMAGE034
is an algorithm. The model and algorithm are different due to different calculation detection components, and the mixed reflection spectrum vector is not changed. The detection index (category) vector, the calculation result and the algorithm and the model matrix are different and different. The hyperspectral oil detection equipment system can establish an algorithm model library according to oil performance, brand, serial number, operating equipment, detection components and detection points. The application software will "connect" (bind) different model algorithms according to the acquisition point. When the detection operation selects an acquisition point, the application software determines which model algorithm needs to be called according to the setting of the reflectivity and the amplitude brightness value obtained by the acquisition operation.
5) And the application software starts multithreading according to different model algorithms, and simultaneously pushes the reflectivity and the amplitude brightness value of the detected oil sample to each thread to perform model calculation. Because each model algorithm is independent, the data volume of the reflectivity and the amplitude brightness value of the uniformly input detected oil sample is limited, and the longitudinal technical independence is formed, so that the parallel operation becomes possible. Therefore, the calculation time cannot be increased because the detection type is increased at a certain acquisition point. For a user, the detection operation of the hyperspectral oil detection equipment is irrelevant to the detection category number, the acquisition detection is clicked on a man-machine interaction page of the equipment, and the detection result can be displayed within 5-6 seconds.
The application software is directed to cross-model detection (multi-model), and when the model is called 1 st time (detection), the model is required to be based on the oil sample background. And self-adaptive learning is performed, so that the detection accuracy is improved. At the 1 st invocation (according to acquisition point) can be attributed to device initialization. The adaptive learning logic is discussed in detail in the next section.
6) And detecting data summarization, and generating a result report implies a distributed calculation and synchronizing the operation process of result data summarization. And the calculation synchronization is to monitor the running of all threads, ensure the end of the operation of the last algorithm model thread and generate a result. And then summarizing the operation results of all threads, and pushing the results to a terminal according to data and report formats. When a certain thread is still calculating, all results are gathered and pushed to a terminal, and the thread result of unfinished calculation is uncertain data or an error result. The delay caused by the synchronization of the operation results of all the threads is microsecond millisecond, and the result display and the operation experience of a terminal client are not influenced.
When a plurality of threads run simultaneously and independently, the starting time and the completion time are different, and it is necessary to ensure that all threads are finished and then the result is taken (or the operation of the next stage is continued) through a synchronous mode (software specific function), otherwise, the result has incompleteness (the result at the moment of software definition may be 'garbage'). Ensuring that the last algorithmic model thread is finished (is random) refers to synchronously monitoring the software functions performed by the threads.
In general, oil samples for device modeling are targeted. The established model is very sensitive to background oil factories, brands, row numbers, viscosity and pollution (particle) degree. The same oil sample collection cross-model detection, such as metal component detection, chemical component detection and oil viscosity detection, cannot ensure that the background information (such as No. 0 oil) of the detected oil sample is matched with the modeling oil sample, and tends to interfere with the model and influence the accuracy. It is necessary to "learn" the model again from the collection points (oil samples) to achieve detection accuracy. The adaptive relearning method is a key link to the implementation of automatic cross-model detection. The method of use employs a folded (subset) interleaved predicted response. The prediction adopts the partial least Square method described by the technical principle of the patent to model and predict (detect), partial subsets are used as observation, the number of the subsets is related by the distribution gradient of the modeled oil sample components, the result response measurement adopts Mean Square Error-MSE,
MSE = (1/k)*Σ(y i – f(x i )) 2 （2）
here, k represents the number of model prediction learning times and is determined by the distribution gradient of oil sample components; y is i Represents the ith observed response; f (x) i ) Represents the ith prediction learning result (detection value). The closer the model learning prediction is to the observed value, the smaller the MSE.
The detection model requires the establishment of a distribution of multiple oil samples. And the oil sample batches provided by the customers are often in a certain area (point) in the oil sample life cycle distribution. Dilution (oil sample), modeling becomes necessary. And (3) selecting a certain 'representative' oil sample from the batch of oil sample group and diluting the oil sample with the base oil (No. 0 oil) to establish a distribution gradient. And forming a modeling oil sample group. Subset (Subset) refers to the sum of prediction and training set in the folding Subset staggered prediction response method, and is smaller than the quantity in the modeling oil sample group depending on the distribution gradient.
The dependence on the distribution gradient (correlation) is because each dilution point (point of the distribution gradient) can be understood as a modeling parameter matrix point. Between the matrix points are successive regions formed by fitting (partial least squares). Therefore, a known oil sample (with test results) and background oil are diluted to form distribution gradient points (training and prediction sets). The results for each point are known. This is used to refine the learning of existing models.
The accuracy of the distribution gradient point (known) depends on the gravimetric-volumetric dilution method.
The present invention uses the following principles to calculate the MSE for a given model:
1. the collected data is split into a training set and a test set according to the model dilution gradient distribution or dependent variable (key detection index elements), as shown in figure 2,
2. the model is refined (machine learning) using only the data in the training set. The test set is predicted (observed) using the model, and the response test MSE is calculated. As shown in fig. 3.
The above steps were repeated k times, each time using a different training set and test set (to model different gradients of the oil sample). According to the training times of the iteration, the predicted value of the model is closer to the output of the training set of the model.
The overall test MSE is calculated as the average of the k test MSEs, as shown in fig. 4.
In practice, we calculate the MSE for a given model using the following procedure:
and (3) diluting the actual collection point oil sample with the laboratory detection result by a volume-specific gravity method in combination with the base oil (No. 0 oil) according to concentration gradient to obtain a group of known distribution oil sample groups. And (3) generating a set of amplitude brightness values and reflectivity by using the oil sample set through hyperspectral oil detection equipment. Wherein each set of intensity values corresponds to a set of reflectivities according to the spectral band,
Figure 39672DEST_PATH_IMAGE035
（3）
Figure 580244DEST_PATH_IMAGE036
is the standard plate amplitude brightness value (obtained when the device is calibrated every day on-line),
Figure 784960DEST_PATH_IMAGE037
is the distribution gradient of the detected oil sample
Figure 905232DEST_PATH_IMAGE038
The amplitude value of the intensity of the position,
Figure 777373DEST_PATH_IMAGE039
is the distribution gradient of the detected oil sample
Figure 957687DEST_PATH_IMAGE038
The reflectivity and the dark current of the position are nothing of hyperspectral oil detection equipmentUnder the irradiation of the light source, the amplitude brightness value (also called as the background noise of the dark room of the equipment, obtained during the daily startup correction of the equipment) is obtained. K is i Is test set, except K i The extrinsic reflectivity and amplitude values are listed as the training set. Dark current
3. Will remove K i The reflectivity and amplitude brightness series are input into the model one by one. The model analyzes the statistical relationship between the dependent variable and the independent variable by adopting a partial least square method. The specific steps and processes are as follows:
a) Residual information matrix
Figure 384121DEST_PATH_IMAGE040
And the component matrix of oil sample (a set of reflectivity and amplitude brightness value) in a certain concentration distribution
Figure 589974DEST_PATH_IMAGE041
Wherein
Figure 300310DEST_PATH_IMAGE040
For the normalized independent variable matrix, each row is a series of constituent indices obtained by dilution, and each column represents a set of band variables corresponding to the training oil sample indices (obtained by dilution).
Figure 402258DEST_PATH_IMAGE041
Is a dependent variable matrix. Also in
Figure 96414DEST_PATH_IMAGE040
Each row is a series of component indices, and each column represents a set of band variables for the corresponding test element index. The data normalization is to subtract the mean of each band and then divide by the standard deviation of each band.
b) Solution matrix
Figure 309220DEST_PATH_IMAGE042
Feature vector corresponding to the maximum feature value of
Figure 139642DEST_PATH_IMAGE043
Calculating a component score vector
Figure 678071DEST_PATH_IMAGE044
And a residual information matrix
Figure 328364DEST_PATH_IMAGE045
Wherein
Figure 610440DEST_PATH_IMAGE046
。
c) Solution matrix
Figure 295369DEST_PATH_IMAGE047
Feature vector corresponding to the maximum feature value of
Figure 739119DEST_PATH_IMAGE048
Calculating a component score vector
Figure 142288DEST_PATH_IMAGE049
And a residual information matrix
Figure 962476DEST_PATH_IMAGE050
Wherein
Figure 501911DEST_PATH_IMAGE051
.
d) Repeating the above steps to the firstmStep, solution matrix
Figure 116563DEST_PATH_IMAGE052
Feature vector corresponding to the maximum feature value of
Figure 7027DEST_PATH_IMAGE053
Calculating a component score vector
Figure 630907DEST_PATH_IMAGE054
。
e) Determining co-extraction based on cross-validationmIndividual component (corresponding wave band of designated element)
Figure 24848DEST_PATH_IMAGE055
Obtaining a satisfactory prediction model and solving
Figure 75981DEST_PATH_IMAGE041
In that
Figure 188162DEST_PATH_IMAGE056
General least squares regression equation above:
Figure 350153DEST_PATH_IMAGE057
(4)
here, theF 0 The variable matrix can also be understood as a label (component) matrix.F m For extraction ofmThe remaining information matrix after each component (label). Beta is a weight parameter, which varies according to different components. Such as beta 1 Represents the component Fe weight parameter. Its spectrum band is 407nm (main spectrum band is beyond 400 nm), beta 1 Will be empirically weighted in the spectral region 400nm-1000nm, at 407 nm.
If data tables X and Y are finally extracted from XmA component of
Figure 129759DEST_PATH_IMAGE058
Substitution into
Figure 351793DEST_PATH_IMAGE059
To obtainpPartial least squares regression equation for each dependent variable:
Figure 685691DEST_PATH_IMAGE060
(5)
here, the
Figure 651373DEST_PATH_IMAGE061
Satisfy the requirements of
Figure 43344DEST_PATH_IMAGE062
，
Figure 436279DEST_PATH_IMAGE063
。
The dependent variable Y is typically the detected indicator (metal component) that needs to be inverted (calculated or recombined). Here Y may be multi-dimensional detection index data. This is the metal content of the oil sample at the point of dilution distribution. And establishing a strict relation between the detected index of the inversion result and the corresponding hyperspectral region, calculating the detection index and diluting the index concentration of the distribution point for calibration, and iteratively converging the calibration to the MSE specified range.
4. Repeating the step "i" times, wherein 1< i < K, K: number of points (starting from 0) of oil sample dilution distribution, each time using a different training set and test set (different gradients of oil sample dilution distribution). According to the training times of the iteration, the predicted value of the model is closer to the output of the training set of the model.
The overall test MSE is calculated as the average of the K test MSEs,
Figure 539364DEST_PATH_IMAGE064
（6）
in general, the more iterations we use in a K-iterative cross-perfect model, the lower the bias of the observation test MSE (difference between the predicted value and the training set of models), but the higher the variance. The "variance" is convolved into the model by the system self-interference (noise). Conversely, the fewer iterations we use, the higher the deviation, but the lower the variance. In practice, selection is performed according to the gradient change of the modeling oil sample and the characteristics of the actual acquisition points. This option has been shown to provide the best balance between bias and variance, thereby providing a reliable estimate of the test MSE, allowing the model to be "learned" again, improving the relative accuracy of detection.
The specific steps of the bulk specific gravity method are as follows:
the method for diluting the oil liquid by simple weighing is adopted, and the oil liquid is diluted according to the preset concentration gradient through the bottom oil and the detection oil sample with the laboratory detection result. The method has the characteristics of simple operation, no need of laboratory equipment and environment, rapidness, accuracy (no accumulated error), no consumption of detection oil sample (and background oil) and the like. The method is an important link which can be implemented by a folding staggered verification method, and not only meets the accuracy of a training set and a test set. The operation process only needs auxiliary equipment: one electronic balance (format: max =200g, e =0.01g, d = 0.001g), two test tubes with a 10ml scale (including a stand that can stand upright), and a pipette. The operation process assumes that the hyperspectral oil detection equipment has a cuvette and a cuvette support which are required for detecting an oil sample. The principle method is as follows,
1. the specific gravity of the base oil and the oil sample with the laboratory test result is obtained. Specific gravity per unit of oil sample can be calculated by injecting 10ml of oil sample into a test tube and subtracting the weight of the test tube from the weight difference.
2. Obtaining a cuvette (3.4 ml) requires diluting the stock solution and measuring the weight of the oil sample. The weight of the oil sample to be injected into the 3.4ml cuvette (2 oil samples mixed) can be deduced from the weight of the two 10ml oil samples (borrowing the results from step 1).
3. And respectively injecting the cuvette bottom oil and the detection oil sample according to the dilution point calculation requirement. The calculation method can be simplified to that,
Figure 26846DEST_PATH_IMAGE065
（7）
Figure 266198DEST_PATH_IMAGE066
the weight of the detected oil sample is injected according to the target dilution concentration,
Figure 79302DEST_PATH_IMAGE067
the target dilution concentration is the dilution point of the oil sample (e.g., diluted from 200ppm to 20ppm relative to the target component) to be detected relative to the laboratory test results, and the assay component is the target component (e.g., fe) in the laboratory test results.
The base oil needs to be added into the cuvette by weight:
Figure 138525DEST_PATH_IMAGE068
。
the cuvette was placed on a support and placed on an electronic balance to obtain the dead weight. And (4) respectively injecting the weights of the detection oil sample and the background oil calculated in the step (3) into the cuvette by using a pipettor. The test oil sample at the dilution point can be reconstructed.
Ships performing ocean-going missions offshore require an effective assessment of the operational health of their diesel engines. On one hand, the cruising ability of the ship is determined, and on the other hand, the operation and maintenance time of landing can be reduced. The timing detection of the metal content, the granularity, the viscosity and the chemical composition of the lubricating oil liquid of the diesel engine is an important index for evaluating the health state of the diesel engine. The engine of a common ship has dozens of oil sample collection points, and the ship is provided with a detection laboratory and a specially-assigned person for operation. The hyperspectral oil detection technology is introduced, so that the detection time can be greatly shortened, the burden of carrying detection equipment is reduced, and arrangement of crews with more effects is realized.
The method is characterized in that background oil (new oil) of different brands (manufacturers) and ranks of the ship diesel engine is collected in each stage, and oil sample samples and changed engine oil samples can be obtained in a directional service oil detection laboratory. And establishing a lubricating oil spectrum model group according to laboratory detection classification (metal components, granularity, viscosity and chemical components) of the lubricating oil of the diesel engine. And inputting each acquisition point through a hyperspectral oil detection software platform. And the acquisition points are classified according to the positions of the engine devices, so that the retrieval is convenient. And binding each acquisition point with the detected class and the spectral model. If a detection range needs to be added to the collection point, for example, detection indexes such as metal component only detection, particle size, viscosity and chemical component addition need to be bound to a model library for particle size, viscosity and chemical component models (models are not established according to the hyperspectral oil detection equipment). The binding method comprises the steps of establishing an oil sample distribution gradient by combining the known detection result oil sample of the acquisition point with background oil through a volume specific gravity dilution method, and carrying out 'training' on a spectral model. The folding staggered verification method is adopted for model training. The method is one-time operation and can be used for initialization before the hyperspectral oil detection equipment is loaded on a ship. Does not hinder the client's real-time operation.
A portable hyperspectral oil detection device (1 kg) implanted with a hyperspectral model group is loaded onto a ship along with a crew, and operation of the hyperspectral model group in the hyperspectral oil detection device is set according to brands (manufacturers) and ranks of lubricating oil in a diesel engine added into maintenance records on the shore. And (4) completing initialization of the hyperspectral oil detection equipment before each sailing.
During navigation, a detected oil sample (cuvette) collected according to regulations is inserted into the hyperspectral oil detection equipment, a collection point is selected through a man-machine interaction page, and the serial number (option, available detection time of the system is used as the serial number) of a container for collecting the oil sample is input. The system may further simplify operations by default acquisition points. The acquisition operation forms a set of reflectance and amplitude value spectra. The system pushes the set of reflectance and amplitude intensity value spectra to a designated model thread (computational unit) according to the acquisition points.
When the algorithm models of all threads (computing units) obtain the computing results, the system collects the computing results and typesets and displays the computing results according to the existing data format. The display mode can be completed through a man-machine interaction interface in the modes of change trend, component indexes, diagnosis and evaluation and the like. All data can be exported, retrieved from the time tag, analyzed, and processed a second time.
The acquired oil sample to be detected passes through an optical system (reflection method), grating light splitting and photoelectric charge conversion after being inserted into hyperspectral oil detection equipment, is calculated and displayed according to the configuration of application scene model classification and a purposeful algorithm model, and the detection of metal components, granularity, viscosity and chemical components of a certain acquisition point of a ship diesel engine at a specified acquisition time point in 5 to 6 seconds is completed. The purposes of simple operation and real-time rapid detection are achieved. The cuvette filled with the oil sample can be stored, traced to the source and reused together with the sampling container.
The invention has the following beneficial effects:
the method can simultaneously obtain the detection results of metal components, granularity, viscosity and chemical components by one-time operation, achieves the effects of simplifying operation, saving consumables, carrying out portable real-time detection and saving the links of operation of professionals.
The word "preferred" is used herein to mean serving as an example, instance, or illustration. Any aspect or design described herein as "preferred" is not necessarily to be construed as advantageous over other aspects or designs. Rather, use of the word "preferred" is intended to present concepts in a concrete fashion. The term "or" as used in this application is intended to mean an inclusive "or" rather than an exclusive "or". That is, unless specified otherwise or clear from context, "X employs A or B" is intended to include either of the permutations as a matter of course. That is, if X employs A; b is used as X; or X employs both A and B, then "X employs A or B" is satisfied in any of the foregoing examples.
Also, although the disclosure has been shown and described with respect to one or an implementation, equivalent alterations and modifications will occur to others skilled in the art based upon a reading and understanding of this specification and the annexed drawings. The present disclosure includes all such modifications and alterations, and is limited only by the scope of the appended claims. In particular regard to the various functions performed by the above described components (e.g., elements, etc.), the terms used to describe such components are intended to correspond, unless otherwise indicated, to any component which performs the specified function of the described component (e.g., that is functionally equivalent), even though not structurally equivalent to the disclosed structure which performs the function in the herein illustrated exemplary implementations of the disclosure. In addition, while a particular feature of the disclosure may have been disclosed with respect to only one of several implementations, such feature may be combined with one or other features of the other implementations as may be desired and advantageous for a given or particular application. Furthermore, to the extent that the terms "includes," has, "" contains, "or variants thereof are used in either the detailed description or the claims, such terms are intended to be inclusive in a manner similar to the term" comprising.
Each functional unit in the embodiments of the present invention may be integrated into one processing module, or each unit may exist alone physically, or a plurality of or more than one unit are integrated into one module. The integrated module can be realized in a hardware mode, and can also be realized in a software functional module mode. The integrated module, if implemented in the form of a software functional module and sold or used as a stand-alone product, may also be stored in a computer readable storage medium. The storage medium mentioned above may be a read-only memory, a magnetic or optical disk, etc. Each apparatus or system described above may execute the storage method in the corresponding method embodiment.
In summary, the above-mentioned embodiment is an implementation manner of the present invention, but the implementation manner of the present invention is not limited by the above-mentioned embodiment, and any other changes, modifications, substitutions, combinations, and simplifications which do not depart from the spirit and principle of the present invention should be regarded as equivalent replacements which are included in the protection scope of the present invention.

Claims (6)

1. A one-time operation detection method for detecting oil sample multi-type indexes is characterized by comprising the following steps:
establishing a relation between oil collecting points and detection indexes, wherein the collecting points are the component content changes of collected oil samples along with the collection time and the oil sample component content changes, and the detection indexes are algorithm models corresponding to hyperspectral oil detection equipment;
the algorithm model server establishes an algorithm model library according to oil performance, brands, ranks, operating equipment, detection components and acquisition points;
pouring the detected oil into a cuvette of a detection instrument from a sampling container, giving serial numbers to the oil sample, and generating the reflectivity and the amplitude brightness value of the oil sample by an optical system of hyperspectral oil detection equipment; binding the reflectivity and the amplitude brightness value with an acquisition point, and uploading the reflectivity and the amplitude brightness value together with the oil sample number and the detection time to an algorithm model server;
selecting different model algorithms according to the acquisition points: when the detection operation selects an acquisition point, determining the reflectivity and amplitude brightness value obtained by the acquisition operation to call one or more model algorithms according to the setting, starting multithreading according to different model algorithms, and simultaneously pushing the reflectivity and the amplitude brightness value of the detected oil sample to each thread to perform model calculation;
detection data summarization: monitoring the running of all threads to ensure that the operation of the last algorithm model thread is finished and the result is generated; and then summarizing the operation results of all threads, and pushing the results to a terminal according to data and report formats.
2. The one-time operation detection method for detecting the multiple types of indexes of the oil sample according to claim 1, wherein when the model is called for detection for the 1 st time, the model is used in advance to carry out self-adaptive learning according to the background of the oil sample; the self-adaptive learning is carried out by adopting a folding subset staggered prediction response method and a partial least square method modeling prediction, namely, the partial least square method modeling prediction is adopted, part of subsets are used as observation, the number of the subsets is related to the distribution gradient of the modeling oil sample components, and the result response is measured by mean square error.
3. The one-time operation detection method for detecting multiple types of indexes of an oil sample according to claim 1, wherein collected data are divided into a training set and a test set according to a modeling dilution gradient distribution or a dependent variable, and the collected data are reflectivity and amplitude brightness values;
training and improving the model only by using data in the training set, predicting the test set by using the model, and calculating a response test Mean Error;
repeating the step K times, wherein K is the number of the interval of the dilution gradient distribution, different training sets and test sets are used each time, and the model enables the predicted value to be close to the output of the model training set according to the iterative training times;
the average of the K test Mean Square Error is taken as the overall test Mean Square Error.
4. The one-time operation detection method for detecting multiple types of indexes of an oil sample according to claim 3, characterized in that an actual collection point oil sample with laboratory detection results is diluted according to concentration gradient by combining a volume-specific gravity method and the bottom oil to obtain a group of oil sample groups with known distribution;
generating a group of amplitude brightness values and a group of reflectivity by using hyperspectral oil detection equipment for an oil sample group, wherein each group of amplitude brightness values corresponds to one group of reflectivity according to a spectrum segment;
will divide test set K i Inputting the reflectivity and amplitude value series into a model one by one, analyzing the statistical relationship between a dependent variable and an independent variable by the model by adopting a partial least square method, wherein the dependent variable Y is the metal component of the oil sample at a certain dilution distribution point;
calculating the detection index and the index concentration of the dilution distribution point for calibration, and iteratively converging to an MSE (mean square error) specified range;
repeating the steps for i times, wherein i is more than 1 and less than K, K is started from 0, different training sets and test sets are used each time to represent different gradients of oil sample dilution distribution, and the model enables the predicted value to be close to the output of the model training set according to the iterative training times.
5. The one-time operation detection method for detecting the oil sample multi-type indexes according to claim 2, wherein the partial least square method comprises the following specific steps:
building a residual information matrixE 0 And a matrix of detected oil sample constituentsF 0 WhereinE 0 For the normalized independent variable matrix, each row is a series of component indexes, and each column represents a group of spectrum variables corresponding to the detection element indexes;F 0 is a dependent variable matrix; also inE 0 Each row is a series of component indices, and each column represents a set of spectral range variables corresponding to a test element index; data normalization, i.e. subtracting the mean value of each spectrum, and then dividing by the standard deviation of each spectrum;
solution matrix
Figure 516954DEST_PATH_IMAGE001
Feature vector corresponding to the maximum feature value ofw 1 Calculating a component score vector
Figure 909889DEST_PATH_IMAGE002
And a residual information matrix
Figure 278554DEST_PATH_IMAGE003
Wherein
Figure 766036DEST_PATH_IMAGE004
；
Solution matrix
Figure 270966DEST_PATH_IMAGE005
Feature vector corresponding to the maximum feature value ofw 2 Calculating a component score vector
Figure 100382DEST_PATH_IMAGE006
And a residual information matrix
Figure 205610DEST_PATH_IMAGE007
In which
Figure 44253DEST_PATH_IMAGE008
；
Repeating the above steps to the firstmStep, solution matrix
Figure 138111DEST_PATH_IMAGE009
Feature vector corresponding to the maximum feature value ofw m Calculating a component score vector
Figure 653275DEST_PATH_IMAGE010
；
Determining co-extraction based on cross-validationmAn ingredientt 1 , t 2 , …, t m Obtaining a satisfactory prediction model, and solvingF 0 In thatt 1 ,t 2 ,…,t m General least squares regression equation above:
Figure 730952DEST_PATH_IMAGE011
wherein the content of the first and second substances,
Figure 107707DEST_PATH_IMAGE012
the weighting parameters of the 1 st, 2 nd and mth components respectively,Fmthe residual information matrix after the m components are extracted;
if data tables X and Y are finally extracted from XmA component of
Figure 587230DEST_PATH_IMAGE013
K =1,2, \ 8230;, m substitution
Figure 273295DEST_PATH_IMAGE014
，Namely obtainpPartial least squares regression equation for individual dependent variables:
Figure 572689DEST_PATH_IMAGE015
here, the
Figure 753135DEST_PATH_IMAGE016
Satisfy the requirement of
Figure 336432DEST_PATH_IMAGE017
，
Figure 678552DEST_PATH_IMAGE018
，
IIs a dependent variablejCorresponding to the parameters of the detected index label,hthe number of dimensions Y, i.e. the number of spectral segments,
Figure 730821DEST_PATH_IMAGE019
is a model matrix parameter, whereinjRepresents the index of the components,nrepresents the index of the spectral band,
Figure 449378DEST_PATH_IMAGE020
is a firstkRelative index of detected componentnResidual information feature vectors for spectral fragments.
6. The single-operation detection method for detecting multiple types of indicators of an oil sample according to claim 4, wherein the bulk specific gravity method comprises the following steps:
calculating the specific gravity of the oil sample by injecting 10ml of oil sample into a test tube through weight difference to obtain the specific gravity of the base oil and the oil sample with laboratory detection results;
obtaining the weight of the cuvette needing to dilute the base oil and detect the oil sample: calculating the weight of 2 oil samples which need to be injected into a 3.4ml cuvette and mixed according to the weights of two 10ml different oil samples;
respectively injecting the weights of the basic bottom oil and the detected oil sample into the cuvette according to the calculation requirements of the dilution point, wherein the calculation method comprises the following steps:
Figure 887182DEST_PATH_IMAGE021
Figure 665782DEST_PATH_IMAGE022
the weight of the detected oil sample is injected according to the target dilution concentration,
Figure 205348DEST_PATH_IMAGE023
the weight of an oil sample to be injected into the cuvette is determined, the target dilution concentration is a dilution point for detecting the oil sample relative to a laboratory detection result, and the detection component is a target component in the laboratory detection result;
the base oil needs to be added into the cuvette by the following weight:
Figure 711284DEST_PATH_IMAGE024
；
placing the cuvette into a support and placing the cuvette on an electronic balance to obtain static weight;
and injecting the detection oil sample and the background oil into the cuvette respectively according to the calculated weight by using a liquid transfer device to obtain the detection oil sample for reconstructing the dilution point.
CN202210852758.2A 2022-07-20 2022-07-20 One-time operation detection method for detecting multiple types of indexes of oil sample Active CN114935548B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210852758.2A CN114935548B (en) 2022-07-20 2022-07-20 One-time operation detection method for detecting multiple types of indexes of oil sample

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210852758.2A CN114935548B (en) 2022-07-20 2022-07-20 One-time operation detection method for detecting multiple types of indexes of oil sample

Publications (2)

Publication Number Publication Date
CN114935548A CN114935548A (en) 2022-08-23
CN114935548B true CN114935548B (en) 2022-11-29

Family

ID=82868204

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210852758.2A Active CN114935548B (en) 2022-07-20 2022-07-20 One-time operation detection method for detecting multiple types of indexes of oil sample

Country Status (1)

Country Link
CN (1) CN114935548B (en)

Family Cites Families (13)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPH0996398A (en) * 1995-09-29 1997-04-08 Shimadzu Corp Lubricating oil managing system
US6897071B2 (en) * 2002-08-13 2005-05-24 Saudi Arabian Oil Company Topological near infrared analysis modeling of petroleum refinery products
US20050197889A1 (en) * 2004-02-11 2005-09-08 Sigma Dynamics, Inc. Method and apparatus for comparison over time of prediction model characteristics
WO2010039295A1 (en) * 2008-10-01 2010-04-08 Chevron U.S.A. Inc. A method for predicting a property of a base oil
US8320637B2 (en) * 2009-08-04 2012-11-27 Chemimage Corporation System and method for hyperspectral imaging of treated fingerprints
CN101923649B (en) * 2010-06-22 2013-01-09 中国海洋大学 Oil spill type recognition method based on fluorescence spectrum
US8849582B2 (en) * 2010-12-21 2014-09-30 Invensys Systems, Inc. Offline analyzer system and method for multivariate characterization of properties in crude and heavy hydrocarbon oils
CN103729650A (en) * 2014-01-17 2014-04-16 华东理工大学 Selection method for near infrared spectrum modeling samples
US9714931B2 (en) * 2014-09-03 2017-07-25 General Electric Company System and method for estimating engine oil health
CN108318446A (en) * 2018-04-13 2018-07-24 山东省花生研究所 It is a kind of detection the total content of vitamin E of peanut seed near-infrared spectroscopy and application
CN110823764A (en) * 2018-08-10 2020-02-21 中国石油化工股份有限公司 Method and device for predicting viscosity of base oil
DE102019213123B3 (en) * 2019-08-30 2021-01-07 Vitesco Technologies GmbH Method and device for determining the motor oil quality of an internal combustion engine
EP3904986B1 (en) * 2020-04-27 2023-08-02 Petroleos del Norte, S.A. Method, system and computer program product for the online monitoring of an oil refinery

Also Published As

Publication number Publication date
CN114935548A (en) 2022-08-23

Similar Documents

Publication Publication Date Title
CN101151623B (en) Method and equipment for classifying sample image features
Kuhn et al. Applied predictive modeling
US5606164A (en) Method and apparatus for biological fluid analyte concentration measurement using generalized distance outlier detection
US5124932A (en) Method for analyzing asymmetric clusters in spectral analysis
CN110687072B (en) Calibration set and verification set selection and modeling method based on spectral similarity
EP2428802B1 (en) Automatic analysis device and analysis method
CA2228844C (en) Biological fluid analysis using distance outlier detection
CN107505346B (en) The method for predicting to be especially the chemical displacement value of NMR spin system in biological fluid sample in class of fluids sample
CN113010389A (en) Training method, fault prediction method, related device and equipment
CN109324013A (en) A method of it is quickly analyzed using Gaussian process regression model building oil property near-infrared
CN108830645A (en) A kind of visitor&#39;s attrition prediction method and system
CN110162811A (en) Testability modeling and analysis method based on fault diagnosis
CN109374631A (en) A kind of tunnel state evaluating method
CN111143981A (en) Virtual test model verification system and method
CN114184599B (en) Single-cell Raman spectrum acquisition number estimation method, data processing method and device
CN114935548B (en) One-time operation detection method for detecting multiple types of indexes of oil sample
Kanoun et al. A framework for dependability benchmarking
JP4366261B2 (en) Method for determining presence / absence of abnormality in measurement reaction process, automatic analyzer capable of executing the method, and storage medium storing program of the method
Ralph et al. An image metric-based ATR performance prediction testbed
CN108416463A (en) A kind of product quality prediction technique and system of hydrocracking process
CN116008216A (en) Method and apparatus for detecting oil blending
JP2023521757A (en) Using a genetic algorithm to determine a model for identifying sample attributes based on Raman spectra
JP2009210445A (en) Production and inspection monitor system, autoanalyzer and data processing method of them
CN105866042A (en) Detection method for spatial distribution of biological quality indexes based on pixel index unbiased estimation method
JP3639721B2 (en) Measuring system

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant