Method for generating estimation model for oil physical properties, method for diagnosing oil physical properties, and system for diagnosing oil physical properties

Abstract

The objective of the present invention is to provide technology for optically quantifying the physical properties of oil.　One aspect of the present invention provides a method for generating an estimation model for physical properties of oil using an information processing system comprising an input device, an output device, a processing device, and a storage device, wherein the information processing system is provided with an estimation model generating unit that generates a machine-trainable estimation model, and the estimation model generating unit accepts input of an absorption spectrum obtained by transmitting light including at least some wavelengths between 800 and 2500 nm through an oil that contains a base oil and an additive, and a value reflecting a physical property value of the oil, and generates the estimation model by machine learning using a value based on the absorption spectrum as an explanatory variable, and the value reflecting the physical property value of the oil as a dependent variable.

Description

Method for generating an estimation model of oil properties, method for diagnosing oil properties, and system for diagnosing oil properties

[0001] The present invention relates to an oil diagnostic technology, particularly to the maintenance of large machinery that uses industrial oils such as lubricating oil, insulating oil, and processing oil, and relates to a technology suitable for monitoring machinery by measuring changes in the physical properties of lubricating oil and other oils that occur with use, and by assessing the remaining life of the oil and predicting the condition of the machinery.

[0002] Diagnosing the physical properties of lubricants used in rotating parts such as bearings and gears is an important technology for maintaining large rotating machinery, such as wind power generator gearboxes, air compressors, ships, power generation turbines, construction machinery, agricultural machinery, cutting machines, pumps, and reduction gears for railway vehicles.

[0003] In addition to lubricating oils, insulating oils are used in transformers and other devices for electrical insulation, and diagnosing the physical properties of insulating oils is also important. Processing oils are also used in machining. There are processing oils for various purposes, such as cutting oil, press oil, heat treatment oil, rust prevention oil, and cleaning oil. In this specification, industrial oils such as lubricating oil, insulating oil, and processing oil are sometimes collectively referred to as oil.

[0004] Lubricating oils are classified into various types depending on their intended use, such as engine oil, turbine oil, hydraulic oil, bearing oil, sliding surface oil, gear oil, compressor oil, and cutting oil. Various additives are blended into the base oil (oil used as the base material) to ensure that each type of lubricating oil meets the required performance. Additives are also blended into other oils to achieve the properties required for each type.

[0005] In recent years, machine condition monitoring has often adopted a strategy to minimize the lifecycle cost of the machine. Large machines such as power generation turbines use large amounts of lubricating oil, and changing the lubricating oil requires stopping the machine, which can result in power generation losses and production stoppages. In addition, costs for purchasing and shipping new oil, oil change work, and waste oil disposal are also required, so it is desirable to use the lubricating oil for as long as possible. Refrigerant fluids in electric vehicles and data centers also undergo oil diagnosis, and replacement or hardware repairs are carried out. Similarly, the color and other characteristics of insulating oil in transformers are monitored.

[0006] Recently, from the perspective of carbon neutrality, automobiles and other vehicles that use large amounts of petroleum-based fuels are being electrified, and fuel demand will decrease in the future. However, there are often no alternatives for industrial oil, so there is a need to minimize its use by extending the oil change interval, etc. This is because reducing oil consumption reduces carbon dioxide emissions. However, overlooking oil deterioration and contamination can lead to machine breakdowns.

[0007] Regarding the physical property diagnosis of lubricating oil, it is necessary to define and distinguish between "deterioration" and "contamination." Broadly speaking, it is necessary to diagnose two types of deterioration: (1) oxidative deterioration of lubricating oil over time, and (2) contamination by external contaminants such as water, dust, and wear particles.

[0008] (1) Oxidative degradation of lubricating oils includes degradation due to oxidation of the base oil and degradation due to the consumption of additives. Oxidative degradation of lubricating oils leads to a decrease in wear resistance, changes in viscosity and viscosity index, a decrease in rust prevention, and a decrease in corrosion prevention. As a result, it can accelerate wear and material fatigue in gearboxes. While it is desirable to use oil for as long as possible, any abnormal deterioration or contamination requires a prompt oil change and equipment inspection.

[0009] As lubricating oil is used for a long time, additives are consumed, resulting in a decline in its performance as a lubricant. Specifically, the consumption of antioxidants can rapidly accelerate the oxidation of base oils. When base oils oxidize, the viscosity of the lubricating oil increases, the thickness of the lubricating film on the sliding surface changes, and proper lubrication performance is no longer achieved. Furthermore, additive consumption is an oxidation reaction, and as additive consumption and base oil oxidation progress, the acid number and base number, indicators of the acidity of the lubricating oil, change. An increase in the acidity of a lubricating oil makes it more susceptible to corrosion and rust on parts such as bearings and gears. Viscosity is sometimes referred to as kinematic viscosity. Changes in density are also observed as the viscosity and acid number of lubricating oils change over time.

[0010] Against this background, efforts have been made to prevent machine breakdowns and avoid excessive oil changes by measuring the viscosity, acid number, base number, and density of lubricating oil.

[0011] Conventional methods for evaluating the physical properties of lubricating oils include definitions for viscosity, acid number, base number, and density. Lubricating oil density represents the mass per unit volume, usually expressed in grams per cubic centimeter. Viscosity (kinematic viscosity) is an index of the fluidity of a lubricating oil. Viscosity is measured by measuring the flow time of a liquid under calibrated gravity. Acid number is the amount of potassium hydroxide required to neutralize the acidic components contained in 1 g of lubricating oil, expressed in mg.

[0012] Base number is an index that indicates the amount of base components contained in engine oil. It is expressed as the number of milligrams of potassium hydroxide, equivalent to the amount of acid, required to neutralize the base components contained in 1 g of lubricating oil. Generally, engine oil contains sulfur, and combustion in the engine produces sulfuric acid, which corrodes cylinders and piston rings. Therefore, basic components are added to engine oil, and the base number of engine oil is defined. Both acid number and base number are measured using neutralization titration.

[0013] All of the conventional lubricant property evaluation methods described above use expensive, precise analytical equipment, which may require the use of reagents or pretreatment during analysis. These methods are evaluated in laboratories or oil analysis companies. On the other hand, installing a sensor in a machine that uses lubricant to learn the oil's properties allows for early detection of abnormalities in the oil and the machine, so oil sensors are sometimes used. Unlike analytical equipment, oil sensors cannot measure all of the oil's properties. Therefore, methods known include MEMS sensors that measure low-viscosity oils, and sensors that measure the permittivity and conductivity of oil to predict ongoing changes in oil properties. Because these sensors require contact with the oil for measurement, the oil sensors are installed by modifying the pipes or tanks through which the oil flows.

[0014] Patent Document 1 describes a technique in which lubricating oil that has deteriorated due to use is filtered through a filter, and the infrared absorption spectrum of the filtered filter is measured by Fourier transform infrared spectroscopy, and the oxidation rate is estimated based on the absorption spectrum. －１ From 4000 cm －１, i.e., mid-infrared spectroscopy in the wavelength range of 2.5 μm to 20 μm is used.

[0015] Patent Document 2 describes a method of extracting organic molecules using a supercritical fluid and measuring the vibrations associated with the C-H bonds of the extracted organic molecules by infrared absorption. ―１ , that is, spectroscopy using mid-infrared light in the wavelength range of 3.125 μm to 6.25 μm is used.

[0016] JP 2016-035400 A JP 2011-102703 A

[0017] There are conventional methods for measuring the viscosity, acid number, base number, and density of lubricating oils using chemical analysis techniques. Recently, there has been an increasing demand for measuring various physical properties of lubricating oils by installing oil sensors on machines that use lubricating oils. However, conventional methods require the application of force to the lubricating oil using equipment or the use of reagents for analysis, making them unsuitable for measurement using oil sensors.

[0018] There is no way to measure the acid value or base value with an oil sensor. Although there is a method to predict them from the dielectric constant and conductivity, it is difficult to predict many phenomena that occur in oil, such as oxidation of the base oil, consumption of additives, the inclusion of foreign matter such as moisture, fine particles, and wear powder, and the generation of sludge, from the dielectric constant and conductivity, and the accuracy is poor.

[0019] There are MEMS type oil sensors that can measure viscosity, but they can only measure low viscosity oils (50 cP or less) such as engine oil, and cannot measure high viscosity oils such as gear oil.

[0020] Installing oil sensors is costly, requires modifying the machine's piping and oil tanks to prevent oil leaks, and poses the challenge of ensuring the reliability of the machine after modification.

[0021] An object of the present invention is to provide a technique for optically quantifying the physical properties of oil.

[0022] One aspect of the present invention is a method for generating an estimation model of oil physical properties, which uses an information processing system including an input device, an output device, a processing device, and a storage device, and the information processing system includes an estimation model generation unit that generates an estimation model that can be subjected to machine learning. The estimation model generation unit receives an absorption spectrum obtained by transmitting light with at least a portion of wavelengths between 800 nm and 2500 nm through an oil containing a base oil and additives, and a value that reflects the physical property values ​​of the oil. The method uses values ​​based on the absorption spectrum as explanatory variables and values ​​that reflect the physical property values ​​of the oil as objective variables to generate an estimation model by machine learning.

[0023] Another aspect of the present invention is a method for diagnosing oil physical properties using a diagnostic system comprising an information processing device that implements the above-described estimation model, the method comprising: inputting into the diagnostic system, as a diagnostic object absorption spectrum, an absorption spectrum obtained by transmitting light of at least a portion of wavelengths between 800 nm and 2500 nm through an oil to be diagnosed of the same type as the oil containing the additive, inputting a value based on the diagnostic object absorption spectrum into the estimation model, and obtaining from the estimation model a value that reflects the physical property values ​​of the oil to be diagnosed.

[0024] Another aspect of the present invention is an oil property diagnostic system including an information processing device that implements the above estimation model.

[0025] According to the present invention, it is possible to provide a technique for optically quantifying the physical properties of oil. Problems, configurations, effects, etc. other than those described above will become clear from the following description of the embodiments.

[0026] Schematic diagram of an optical sensor that measures near-infrared absorption spectra. Schematic diagram of another optical sensor that measures near-infrared absorption spectra. Graph showing the relationship between kinematic viscosity and temperature of oil in an embodiment. Graph showing the relationship between oil density and temperature in an embodiment. Graph showing the relationship between machine operating time and oil viscosity, acid number, base number, and insoluble matter in an embodiment. Block diagram of an oil diagnostic system in an embodiment. Flowchart of an oil diagnostic method in an embodiment. Flowchart of an actual measurement data collection process in an embodiment. Graph of an example of a near-infrared spectrum of engine oil. Flowchart of a material state estimation model generation process in an embodiment. Graph of an example of a near-infrared spectrum differentiated twice. Flowchart of an oil physical property estimation process in an embodiment. Graph showing the consistency between predicted values ​​and actual measured values ​​using an estimation model in an embodiment. Flowchart showing an example of engine oil viscosity estimation taking into account the temperature characteristics of viscosity in an embodiment. Schematic diagram of a lubricant monitoring system for a wind turbine generator in an embodiment. Conceptual diagram of a rotating part equipped with a lubricant sensor. Flowchart showing a lubricant diagnostic process in an embodiment. Graph showing an example of displaying results.

[0027] The embodiments will be described in detail with reference to the drawings. However, the present invention should not be interpreted as being limited to the description of the embodiments shown below. Those skilled in the art will easily understand that the specific configuration can be changed within the scope of the idea or purpose of the present invention.

[0028] In the configurations of the embodiments described below, the same parts or parts having similar functions are denoted by the same reference numerals in different drawings, and redundant explanations may be omitted.

[0029] When there are multiple elements with the same or similar functions, they may be described using the same reference numeral with different subscripts. However, when there is no need to distinguish between multiple elements, the subscripts may be omitted.

[0030] The terms "first," "second," "third," etc. used in this specification are used to identify components and do not necessarily limit the number, order, or content of the components. Furthermore, numbers used to identify components are used in different contexts, and numbers used in one context do not necessarily indicate the same configuration in another context. Furthermore, this does not prevent a component identified by a certain number from also serving the function of a component identified by another number.

[0031] In order to facilitate understanding of the invention, the position, size, shape, range, etc. of each component shown in the drawings, etc. may not represent the actual position, size, shape, range, etc. Therefore, the present invention is not necessarily limited to the position, size, shape, range, etc. disclosed in the drawings, etc.

[0032] All publications, patents, and patent applications cited herein are hereby incorporated by reference in their entirety.

[0033] Elements referred to in the singular herein include the plural unless the context clearly indicates otherwise.

[0034] In this specification, the wavelength range from 800 nm to 2500 nm is referred to as the near-infrared region.

[0035] In the embodiment, a machine monitoring method using changes in optical absorption intensity in the near-infrared region of oils containing base oils and additives, such as lubricating oil, diagnoses oil deterioration and contamination using a near-infrared optical sensor. In principle, transmittance = 1 - absorptance. In the embodiment, transmittance and absorptance have the same meaning. Transmittance and absorptance can also be expressed as percentages. Absorbance, also known as optical density, is the negative value of the common logarithm of transmittance.

[0036] Lubricating oils include engine oil, turbine oil, hydraulic oil, bearing oil, sliding surface oil, gear oil, compressor oil, and cutting oil. Lubricating oils are composed of base oil and additives. Additives include antioxidants, rust inhibitors, antifoaming agents, viscosity index improvers, oiliness improvers, extreme pressure additives, detergents and dispersants, pour point depressants, and emulsifiers. Oils used for purposes other than lubrication include transformer oil, cleaning oil, and cutting oil. For many oils, the degree of additive consumption through use can be used to determine when to change the oil or to indicate any abnormalities in the machine.

[0037] One example described in the examples is a diagnostic method for oil containing base oil and additives. The method measures the intensity of light transmitted through the oil using an optical sensor having a detector sensitive to wavelengths in the range of 800 nm to 2500 nm, i.e., the near-infrared region, and determines the viscosity, acid number, and base number of the oil in advance using a method such as chemical analysis. The method then determines the viscosity, acid number, and base number of the oil using machine learning.

[0038] According to this configuration, when detecting changes in the physical properties of oil (measurement object) by measuring changes in the light absorption intensity of the oil in the wavelength range of 2500 nm to 800 nm using an optical sensor, the state of the oil can be accurately determined.

[0039] One example of an embodiment is a method for generating an estimation model of oil physical properties, which uses an information processing system including an input device, an output device, a processing device, and a storage device, and the information processing system includes an estimation model generation unit that generates an estimation model capable of machine learning, and inputs an absorption spectrum obtained by transmitting light of at least a portion of wavelengths between 800 nm and 2500 nm through an oil containing a base oil and additives, and a value reflecting the physical properties of the oil, into the estimation model generation unit, and generates an estimation model using values ​​based on the absorption spectrum as explanatory variables and values ​​reflecting the physical properties of the oil as objective variables.

[0040] Another aspect of the present invention is a method for diagnosing oil properties using a diagnostic system consisting of an information processing device that implements the above-mentioned estimation model, in which an absorption spectrum obtained by transmitting light with at least a portion of wavelengths between 800 nm and 3000 nm through an oil of the same type as the oil to be diagnosed, which contains the base oil and the additives, is input to the diagnostic system, a value based on the absorption spectrum is input to the estimation model, and a value reflecting the physical properties of the oil to be diagnosed is obtained from the estimation model.

[0041] Another example is an oil property diagnostic system that includes an information processing device that implements the above estimation model.

[0042] Another example is a method for diagnosing oil properties, which comprises using an optical sensor having a light source and a detector for detecting light emitted from the light source to obtain an absorption spectrum in the wavelength range of 800 nm to 2500 nm, inputting data based on the absorption spectrum into an estimation model implemented in an information processing device, and predicting the concentration of the additive.

[0043] <Method of measurement by sensor> Figure 1 is a schematic diagram showing the configuration of an example of an optical sensor. The near-infrared absorption spectrum measurement of a lubricant is performed by an optical sensor 700 having a light source that generates light with a wavelength in the near-infrared region and a detector that detects wavelengths in the near-infrared region. The lubricant to be measured is measured by a detector 103, which detects the light intensity of light emitted from a light source 101 (indicated by an arrow in the figure) after it passes through a lubricant 102.

[0044] FIG. 1 shown above shows one form of sensor, but it is also possible to have the light emitted from the light source 101 pass through the lubricating oil 102 and enter the detector 103 using means such as reflection by a mirror, focusing by a lens, or passing through a prism.

[0045] 2 is a diagram showing another embodiment of the optical sensor. In the optical sensor 700A, a light source 101 and a detector 103 are arranged on the same plane and in contact with the lubricating oil 102, with a reflector 104 installed on the opposite side, and light emitted from the light source 101 (indicated by the arrow in the figure) is reflected by the reflector 104 and received by the detector 103. Even in this embodiment, a lens, a mirror, or a prism may be installed along the optical path.

[0046] By using this type of measurement, absorption spectrum data can be obtained without bringing the optical sensor into contact with the oil.

[0047] <Scope of application> Types of oils containing base oil and additives include engine oil, turbine oil, hydraulic oil, bearing oil, slideway oil, gear oil, compressor oil, cutting oil, insulating oil, cutting processing oil, press processing oil, heat treatment oil, rust preventative oil, cleaning oil, etc. Types of additives include antioxidants, rust inhibitors, antifoaming agents, viscosity index improvers, oiliness improvers, extreme pressure additives, detergent dispersants, pour point depressants, emulsifiers, etc.

[0048] <Near-infrared spectroscopy> Near-infrared spectroscopy is an analytical method using light with wavelengths ranging from 800 nm to 2500 nm. Compared to ultraviolet-visible spectroscopy, which has shorter wavelengths than near-infrared, and mid-infrared spectroscopy, which has longer wavelengths than near-infrared, near-infrared wavelengths are characterized by wavelengths in the ultraviolet-visible region where electronic transitions occur. Light absorption occurs primarily due to functional groups such as aromatic benzene rings and naphthalene rings, making it difficult to obtain detailed information about molecular structure. Mid-infrared wavelengths are characterized by absorption due to intramolecular electronic transitions, i.e., intramolecular interatomic vibrations (stretching vibrations, bending vibrations, etc.) such as C-H and O-H. While a great deal of information about molecular structure can be obtained, the direct transitions result in strong absorption and a tendency for absorbance saturation, making it difficult to predict physical properties from mid-infrared absorption spectra. In the near-infrared region, which is the wavelength range between ultraviolet-visible and mid-infrared, there is almost no absorption due to electronic transitions, but rather absorption occurs at wavelengths equivalent to the sum of the absorption energy due to multiple intramolecular transitions, known as energy transitions (overtones) or combination tones, which are two or three times higher than absorption in the near-infrared region.

[0049] Quantitative analysis using optical absorption spectroscopy generally uses the Beer-Lambert law, which states that absorbance is proportional to the concentration and the optical path length. The relationship between transmittance and absorbance is determined by taking the common logarithm of the transmittance, and the negative value is taken as the absorbance.

[0050] Since absorption in the near-infrared region is a forbidden transition, it is much weaker than absorption in the mid-infrared region, and the Beer-Lambert law can be applied very well to absorbance measurements using a transparent cell with an optical path length of approximately 3 mm to 20 mm.

[0051] <Physical property prediction using near-infrared absorption spectrum data> Almost all base oils and additives that make up lubricating oils are organic compounds. Absorption spectra in the near-infrared region contain a wealth of information about the molecular structure of the organic compounds that make up the lubricating oil. Meanwhile, the inventors discovered that physical properties such as viscosity, acid number, and base number of lubricating oils are determined by the molecular structure that makes up the lubricating oil. Therefore, the inventors discovered that it is possible to predict physical properties such as viscosity, acid number, and base number of lubricating oils from near-infrared absorption spectrum data, which contains a wealth of information about the molecular structure that makes up the lubricating oil. Lubricating oil physical properties can be predicted using the following procedure.

[0052] When there are multiple lubricant samples with different usage times for a single lubricant, the viscosity, acid number, and base number of each sample are measured using an analytical device, etc. Near-infrared absorption spectrum data is obtained for each sample using a spectrophotometer or optical sensor capable of measuring absorbance in the near-infrared region.

[0053] Spectral preprocessing is an effective step in spectral analysis. In the first step, smoothing is performed to remove spectral noise. For example, a method is used in which the centered moving average method is applied to a selected interval of approximately 3 to 20 adjacent points. In the second step, a second-order derivative filter is applied to the spectrum, which makes it possible to extract wavelengths with large spectral changes and extract peaks from spectra with complex overlaps and unclear peak positions, which are characteristic of near-infrared absorption spectra. Either of these two steps can be performed first. For the derivative filter, a first-order derivative can be used, or it is possible to proceed with the analysis without a derivative filter.

[0054] Next, a model for predicting viscosity, acid number, and base number from near-infrared absorption spectrum data can be created by applying a multivariate analysis method such as Partial Least Squares Regression (PLS), with viscosity, acid number, and base number as the response variables and preprocessed near-infrared absorption spectrum data as the explanatory variables. Validation, which is performed after the prediction calculation, is the process of verifying the input data and model parameters so that the model or system outputs accurate and reliable results, and is selected from the following common methods.

[0055] Cross-validation can be used when the number of samples is not large, as in big data. k-fold cross-validation divides a dataset evenly into k folds, and in each evaluation, one fold is used as the test set and the remaining k-1 folds are used as the training set. Leave-One-Out Cross-Validation (LOOCV) uses each sample in the dataset as a test set and evaluates the remaining samples as the training set, which is useful when the amount of data is small. Stratified k-fold cross-validation divides the data into k folds, as in k-fold cross-validation, but ensures that the class proportions within each fold are the same as in the original dataset. Either method can be used, but when comparing results, the same validation method should be used.

[0056] By creating such a model, for example, by measuring the near-infrared absorption spectrum of a lubricant sample with unknown viscosity, the viscosity can be predicted using the model. However, such a model is only applicable to the same lubricant; for lubricants with different compositions, different lubricant products from the same manufacturer, or lubricants from different manufacturers, a separate model must be created.

[0057] <Predicting viscosity using near-infrared absorption spectrum data> Viscosity has temperature dependency, or temperature characteristics. Viscosity μ can be expressed as a function of temperature T: μ = f(T) Figure 3 is an example graph showing the relationship between oil kinematic viscosity and temperature. As shown in Figure 3, the higher the temperature, the easier the molecules in the lubricating oil move, and therefore the lower the viscosity. Since the near-infrared absorption spectrum contains a lot of information about the molecules that make up the lubricating oil, it is possible to predict the viscosity from the near-infrared absorption spectrum of the lubricating oil.

[0058] Lubricants are sometimes evaluated using a physical property called kinematic viscosity ν. Viscosity μ indicates the difficulty of an object moving in a fluid. On the other hand, kinetic viscosity ν indicates the difficulty of the fluid itself to move. The density of an object affects this difficulty of movement. Even if the viscosity μ is the same, if the density is different, the difficulty of movement (kinetic viscosity ν) will be different. Kinematic viscosity ν can be obtained by dividing viscosity μ by density.

[0059] The procedure for predicting viscosity μ from near-infrared absorption spectra is as follows. As mentioned above, viscosity μ is known to have temperature characteristics. Furthermore, the temperature of lubricating oils used in machinery changes from moment to moment depending on the outside air temperature and operating conditions. If viscosity μ has temperature characteristics, this means that the near-infrared absorption spectrum of the lubricating oil also has temperature characteristics. When continuously measuring viscosity μ to track changes in the lubricating oil over time, it is necessary to determine in advance that viscosity μ will be obtained at a certain temperature, otherwise changes in the lubricating oil cannot be tracked.

[0060] Therefore, since the temperature characteristics of the viscosity μ of a lubricating oil are determined depending on the type of lubricating oil, and the viscosity μ can be expressed as a function of temperature, it is possible to obtain the near-infrared absorption spectrum at a certain time and at the same time obtain the measured temperature of the lubricating oil, and then use the temperature characteristic formula for viscosity μ, which has been experimentally obtained in advance, to obtain the viscosity μ at a reference temperature, relative to the viscosity μ predicted from the obtained near-infrared absorption spectrum.

[0061] To achieve high accuracy, it is advisable to determine the temperature characteristic equation of viscosity μ (temperature characteristic information of viscosity) for each type of lubricant (type of base oil and type of additive). It may also be determined for each usage time of the lubricant. However, in order to sacrifice accuracy for efficiency, one or more generalized temperature characteristic information may be used for multiple types of lubricant.

[0062] The "kinematic viscosity ν" is sometimes used to evaluate the performance of lubricating oils. As mentioned above, kinematic viscosity ν is the value obtained by dividing viscosity μ by density, and density also has temperature characteristics. Therefore, by creating a temperature correction formula for the density of the lubricating oil to be measured in advance, it is possible to determine the density of the lubricating oil from the temperature at which the near-infrared absorption spectrum is measured, and then divide the viscosity μ obtained from the near-infrared absorption spectrum by the density to determine the kinematic viscosity ν.

[0063] To achieve high accuracy, it is advisable to determine the density-temperature characteristic formula (density-temperature characteristic information) for each type of lubricant (type of base oil and type of additive). It may also be determined for each usage time of the lubricant. However, to sacrifice accuracy for efficiency, one or more generalized temperature characteristic information may be used for multiple types of lubricant.

[0064] Figure 4 shows an example of the relationship between temperature and density of a lubricating oil product. There are various grades of lubricating oil products with different viscosities, but in general, the density of lubricating oil tends to decrease as the temperature increases.

[0065] Since the acid value and base value have no temperature characteristics, there is no need to perform temperature correction on the near-infrared absorption spectrum.

[0066] <Relationship between machine operating time and viscosity, acid number, and base number> Figure 5 is a graph showing the relationship between the operating time of a gasoline engine and each of the engine oil's properties: viscosity, acid number, base number, and insoluble content. Operating time can be evaluated by converting it to mileage. The viscosity of engine oil decreases for a while after the start of use and then gradually increases. The initial decrease in viscosity is due to molecules being broken down by shear force. The acid number of engine oil gradually increases after the start of use, and the base number continues to decrease from the beginning.

[0067] In Figure 5, after 11,000 hours, oxidation of the oil progresses, and the insoluble matter that causes sludge and varnish increases. It is desirable to determine whether to change the oil before the insoluble matter begins to increase. Viscosity, acid number, and base number can be predicted from near-infrared absorption spectra. Continuous measurement of any of these parameters is effective for predicting oil life. Continuous measurement of multiple parameters allows for highly accurate predictions. Because viscosity measurements may vary depending on the temperature, it is necessary to compare viscosities at the same temperature; for example, the viscosity at 40°C. Other temperatures within the oil's operating range are also acceptable. Alternatively, kinematic viscosity, which takes into account the density of the oil, may be used instead of viscosity.

[0068] 6 shows an example of a system used for diagnosing engine oil in the embodiment. The diagnostic system 1200 in the embodiment is configured using a normal computer such as a server. The diagnostic system 1200 includes an input device 1210, an output device 1220, and a processing device 1230, and stores programs and data for processing in a storage device 1240. The storage device 1240 may be configured by combining known storage devices such as semiconductor memories and magnetic disk devices.

[0069] The storage device 1240 includes an estimation model generation unit 1250 and an oil property estimation unit 1260. In this embodiment, the estimation model generation unit 1250 and the oil property estimation unit 1260 are included in the same device, but they may also be configured as separate devices.

[0070] The estimation model generation unit 1250 includes a measurement database 1251, a preprocessing unit 1252A, a learning database 1253, and a multivariate analysis unit 1254. The oil property estimation unit 1260 includes a preprocessing unit 1252B, an estimation model 1261, and a temperature compensation unit 1262.

[0071] 7 shows an example of engine oil diagnosis using the diagnosis system 1200. First, measurement data is acquired from actual engine oil, and a measurement database 1251 is generated. This is performed by collecting engine oil samples and analyzing the engine oil using known methods (S1310). An example of the measurement data is shown in FIG. 5.

[0072] A preprocessing unit 1252A of the estimation model generation unit 1250 prepares explanatory variables and objective variables for model generation from the actual measurement data and stores them in the learning database 1253. A multivariate analysis unit 1254 generates an estimation model 1261 of the oil properties using the explanatory variables and objective variables (S1320).

[0073] The oil property estimation unit 1260 predicts the oil property using the obtained estimation model 1261 (S1330). Using the prediction result, for example, if the viscosity of new engine oil is 20 cP, it is recommended to change the oil when the viscosity reaches 25 cP through use.

[0074] FIG. 8 is a flow diagram of the actual measurement data collection process S1310. The above engine oil was applied to an automobile engine and continuously operated. Every 10 hours (S1311), 10 ml of engine oil was sampled (S1312), and the near-infrared absorption spectrum was measured using the optical sensor shown in FIG. 1 or 2 (S1313). The wavelength resolution of the near-infrared absorption spectrum was 2 nm. The viscosity, acid number, and base number of the engine oil were measured using any known method (S1314). Although not required, color information of the engine oil may also be measured if necessary (S1315). The measured near-infrared absorption spectrum, antioxidant concentration, viscosity, acid number, and base number are linked to time information and recorded in the actual measurement database 1251 (S1316). Note that the sample volume and sample acquisition time interval are merely examples. Furthermore, the oil physical properties and analytical methods for these properties may also be selected from known methods.

[0075] Figure 9 shows an example of the near-infrared absorption spectrum of engine oil. In this example, the absorbance is shown for wavelengths from 1550 nm to 1950 nm. The curves show the spectra after different periods of use.

[0076] Next, explanatory variables are generated using the measured near-infrared absorption spectrum, and a model for estimating the physical properties of the oil is generated using the separately measured viscosity as the response variable (FIG. 7, S1320).

[0077] In generating the model, it is possible to use a known machine learning method in which explanatory variables are input and a target variable is output. In this example, a PLS regression analysis was performed.

[0078] 10 shows a flow diagram of the oil property estimation model generation process S1320. This process is performed by the estimation model generation unit 1250, which is a general computer, and the processing unit 1230 executes software.

[0079] First, the preprocessing unit 1252A reads out data of the near-infrared absorption spectrum at a certain time point (S1321) from the actual measurement database 1251. The preprocessing of the near-infrared absorption spectrum in the preprocessing unit 1252A includes smoothing by the Savitzky-Golay (SG) method using data from nine adjacent points (S1322) and second-order differentiation of the spectrum (S1323).

[0080] Figure 11 shows a graph of the results of differentiating the spectrum twice. The curves represent the derivatives of the spectrum at different times after use. Differentiation can emphasize the differences in the near-infrared absorption spectrum over time. Differentiation may not be necessary in some cases. Differentiation may also be performed only once. Alternatively, differentiation may be performed three or more times.

[0081] The twice-differentiated near-infrared absorption spectrum is stored in the learning database 1253 as an explanatory variable x, and the antioxidant concentration at the same time as the objective variable y.

[0082] The multivariate analysis unit 1254 performed PLS regression analysis using the explanatory variable x and the objective variable y (S1324) to generate an oil property estimation model 1261. After the PLS regression analysis, cross-validation was performed (S1325). The generated estimation model 1261 is implemented in the oil property estimation unit 1260.

[0083] 12 is a flow diagram of the oil property estimation process S1330 using the generated estimation model 1261. This process is performed by a general computer-based oil property estimation unit 1260 through software processing.

[0084] First, the near-infrared absorption spectrum of the engine oil whose physical properties are to be determined is acquired and input from the input device 1210 of the diagnostic system 1200 (S1331). The near-infrared absorption spectrum is acquired by the optical sensor shown in Figure 1 or Figure 2. One of the features of the embodiment is that the near-infrared absorption spectrum can be collected contactlessly and remotely.

[0085] The near-infrared absorption spectrum is subjected to preprocessing such as smoothing and second differentiation by the preprocessing unit 1252B in the same manner as when the estimation model was generated in Fig. 10 (S1332). The preprocessed near-infrared absorption spectrum is input to the oil property estimation model to estimate the viscosity (S1333).

[0086] The near-infrared absorption spectrum of the engine oil was measured once a day, and when the predicted viscosity reached 25 cP, a notification was sent and the engine oil was changed.

[0087] Figure 13 is a graph showing the consistency between predicted and measured viscosity values ​​using the estimation model 1261. The reference values ​​plotted as triangles are the measured values ​​of the samples, and the result of analysis using the PLS model is the PLS regression equation shown as a straight line. The results predicted from the measured values ​​using this regression equation (estimation model) are the predicted values, and if the reference values ​​and the predicted values ​​match well, a good PLS model has been constructed. Using the PLS regression equation in Figure 13 as a calibration curve, the viscosity of a sample with unknown viscosity can be quantified.

[0088] In the above example, the viscosity of the engine oil is finally determined and output from the output device 1220, but as will be described later, the acid number, base number, or usage time may also be output. Furthermore, the estimation model was generated using multivariate analysis using software capable of executing PLS without using special hardware, but other well-known machine learning methods using a GPU (Graphic Processor Unit) or the like may also be employed.

[0089] FIG. 14 shows an example of engine oil viscosity estimation that takes into account the temperature characteristics of viscosity for more accurate viscosity prediction.

[0090] To measure the near-infrared absorption spectrum of engine oil, a sensor is attached to the engine, or engine oil is sampled periodically, such as once a week, and the near-infrared absorption spectrum of the engine oil is measured. Assume that the temperature of the lubricating oil when the near-infrared absorption spectrum is acquired is 70°C. The acquired spectral data is input to preprocessing unit 1252B (S2001), and second-order differentiation and smoothing are performed under appropriate conditions as preprocessing (S2002).

[0091] The estimation model 1261 performs PLS regression analysis on the preprocessed data to obtain a predicted viscosity value (S2003). The predicted viscosity value output by the estimation model 1261 is a predicted viscosity value at 70° C. Thereafter, the temperature compensation unit 1262 performs temperature compensation.

[0092] The temperature at which the near-infrared absorption spectrum is measured is acquired (S2004), and this temperature value and the temperature characteristic equation for the viscosity of the lubricating oil (S2005) are used to estimate the viscosity at 40°C from the predicted viscosity value at 70°C (S2006).

[0093] Furthermore, using the separately obtained temperature characteristic equation for density (S2007), the predicted viscosity value at 40° C. is divided by the density value at 40° C. to estimate the kinematic viscosity at 40° C. Here, an example of predicted viscosity and kinematic viscosity at 40° C. is shown, but it is also possible to predict kinematic viscosity at, for example, 60° C. or 100° C. as long as it is within a range that can be covered by the temperature characteristic equation for viscosity and density.

[0094] The temperature characteristic equation showing the temperature dependency of viscosity (or kinematic viscosity) and the temperature characteristic equation (or table) showing the temperature dependency of density as described in FIG. 3 are obtained in advance by actual measurement, simulation, or as theoretical values, and stored.

[0095] An example of diagnosing hydraulic oil (hereafter referred to as hydraulic oil) on a large ship is shown below using the system in Figure 6. The acid number of new hydraulic oil was 1.6, and it was supposed to be changed when the acid number reached 2. From past experience, it was known that the oil would need to be changed after an average of 3,000 hours of use.

[0096] The hydraulic oil was continuously used under the same conditions as those used on a large ship, and 15 ml was sampled every 10 hours. The near-infrared absorption spectrum of the sampled oil was obtained using the optical sensor shown in Figure 1. The acid value, viscosity, and contamination level (mass method) were also measured using the sampled oil.

[0097] The obtained near-infrared absorption spectrum was used as the explanatory variable in PLS regression analysis, and the acid value was used as the target variable. The near-infrared absorption spectrum was smoothed at 11 points and differentiated twice, and then PLS regression analysis was performed. Cross-validation was performed on the results. The results of this analysis demonstrated that it is possible to predict the acid value from the near-infrared absorption spectrum.

[0098] Furthermore, for this hydraulic oil, viscosity was used as the objective variable, and near-infrared absorption spectrum data was used as the explanatory variable. After performing nine-point smoothing and double differentiation on the near-infrared absorption spectrum, PLS regression analysis was performed. Cross-validation was performed on the results, and it was found that viscosity can be predicted from the near-infrared absorption spectrum of hydraulic oil.

[0099] An example of gas engine oil diagnosis is shown below using the system shown in Figure 6. The viscosity of new oil is 10 cP, and it is recommended to change it when the viscosity drops to 13 cP. The base number of the new oil is also 50, and it is recommended to change it when it drops to 20.

[0100] This gas engine was operated continuously, and an optical sensor capable of measuring near-infrared absorption spectra was installed in a sight glass (a part made of a transparent material that is installed so that the color of the oil can be seen) installed in the engine oil piping.

[0101] Near-infrared absorption spectra were measured every hour during engine operation, continuing up to 1000 hours. In addition, small amounts of engine oil were sampled every 20 hours, and viscosity and base number measurements were performed.

[0102] A PLS regression analysis was performed using the acquired near-infrared absorption spectrum and separately measured viscosity data to generate a viscosity estimation model. Using the estimation model, it was confirmed that viscosity can be predicted from the near-infrared absorption spectrum of gas engine oil. Using a similar method, it was also confirmed that base number can be predicted from the near-infrared absorption spectrum of gas engine oil.

[0103] In this embodiment, the configurations of the first to third embodiments are applied to a system and method for monitoring lubricating oil for a wind power generator. This embodiment is a system for monitoring lubricating oil supplied to a mechanical drive unit of a wind power generator. This system includes a diagnostic system 1200 shown in FIG. 6.

[0104] The storage device in the monitoring system stores additive concentration data that stores the concentration of additives in the lubricating oil in chronological order as a reference, and the diagnostic system 1200 estimates the time when the viscosity or acid value of the lubricating oil, determined from the near-infrared absorption spectrum of the lubricating oil, will reach a predetermined threshold value.

[0105] (1. Overall System Configuration) Figure 15 shows a schematic diagram of a lubricant monitoring system for a wind power generator having a lubricant supply system. Inside the nacelle 3 of the wind power generator 1, there are a main shaft 31, a gearbox 33, a generator 34, and bearings such as yaw and pitch bearings (not shown), all of which are supplied with lubricant from an oil tank 37. The system also has the general components of a wind power generator, such as a hub 4, a nacelle bulkhead 30, a shrink disk 32, a main frame 35, a radiator 36, and a coupling 38.

[0106] 15, multiple wind turbines 1 are usually installed on the same site, and these are collectively referred to as a farm 200a, etc. Each wind turbine 1 is equipped with various sensors (not shown) in its lubricating oil supply system, and sensor signals reflecting the state of the lubricating oil are collected in a server 210 in the nacelle 3.

[0107] Furthermore, sensor signals obtained from the server 210 of each wind turbine generator 1 are sent to an aggregation server 220 arranged for each farm. Data from the aggregation server 220 is sent to a central server 240 via a network 230. Data from other farms 200b and 200c is also sent to the central server 240. Furthermore, the central server 240 can send instructions to each wind turbine generator 1 via the aggregation server 220 and server 210. The central server 240 basically has the functions of the diagnostic system 1200 shown in Figure 6, and the system of this embodiment is capable of remote oil monitoring.

[0108] (2. Sensor Arrangement) Fig. 16 is a conceptual diagram of a rotating part equipped with a lubricant sensor. Lubricant is supplied to a rotating part 302 from a lubricant supply device 301 such as a pump. The lubricant supply device 301 is connected to an oil tank 37 and receives the supply of lubricant. The rotating part 302 is, for example, a gearbox 33 or any other general part that comes into mechanical contact, and is not particularly limited.

[0109] The optical sensor 304 is disposed in the lubricating oil flow path etc. in order to detect the state of the lubricating oil. A concrete example of the optical sensor 304 is shown in FIG.

[0110] In this embodiment, a transparent measurement unit 303 is provided in a flow path (near the end of the lubricant path) branching off from the lubricant flow path connected to the lubricant drain port of the rotating part 302, and a portion of the lubricant is introduced into this measurement unit 303. An optical sensor 304 is then installed in the measurement unit 303. The measurement unit 303 is not provided in the main lubricant flow path in order to adjust the flow rate of the lubricant in the measurement unit 303 to a flow rate suitable for detecting the state of the lubricant. The lubricant discharged from the rotating part 302 returns to the oil tank 37 via a filter 305. Note that the filter 305 is not essential. The optical sensor 304 measures the near-infrared absorption spectrum of the oil. The state of the lubricant can be evaluated based on the temporal change in the near-infrared absorption spectrum of the lubricant.

[0111] In this embodiment, the optical sensor 304 includes an optical sensor equipped with a near-infrared light source and a light receiving element, and acquires the near-infrared absorption spectrum of the lubricating oil.

[0112] The optical sensor 304 transmits near-infrared light from a near-infrared light source through the oil and detects the near-infrared light that has passed through the oil with a light-receiving element with sensitivity according to the wavelength, thereby acquiring a near-infrared absorption spectrum. The acquired near-infrared absorption spectrum reflects the oil's absorptivity or transmittance for light of each wavelength. The acquired near-infrared absorption spectrum and an estimation model are used to determine the viscosity and acid value of the lubricating oil, and to perform deterioration diagnosis and remaining life diagnosis.

[0113] The quality of lubricating oil deteriorates with use and it no longer performs its original function. For this reason, maintenance such as replacement is required depending on the degree of quality deterioration. In order to know the timing of such maintenance, it is useful for the efficiency of maintenance management to be able to remotely monitor data that can be collected by optical sensors 304 installed on-site. The data collected by the optical sensors 304 is collected, for example, in a server 210 in the nacelle 3, and then sent via an aggregation server 220 that aggregates data within the farm 200 to a central server 240 that aggregates data from multiple farms.

[0114] However, for analyses that require equipment for measurement, such as viscosity measurement and acid number measurement specified in ASTM standards, it is necessary to collect lubricating oil samples as appropriate and analyze them using separately provided equipment. The results of these viscosity and acid number measurements are also stored as data in the central server 240, and it is desirable to aggregate the data and take this data into consideration when understanding the physical properties of the lubricating oil.

[0115] The aggregated data may include not only data related to lubricants but also data indicating the operating status of the wind turbine. For example, the wind turbine output value (the higher the value, the faster the lubricant deteriorates), the actual operating time (the longer the value, the faster the lubricant deteriorates), the machine temperature (the higher the value, the faster the lubricant deteriorates), the shaft rotation speed (the faster the lubricant deteriorates), etc. These data can be collected from sensors with known configurations installed at various locations on the wind turbine or from control signals from the device.

[0116] (3. Lubricant Diagnosis Flow) Figure 17 is a flow diagram showing the lubricant diagnosis process according to this embodiment. The process shown in Figure 17 is performed under the control of any of the server 210, aggregation server 220, or central server 240 in Figure 15. In the following example, it is assumed that the process is performed by the central server 240. Functions such as calculation and control are realized by software stored in the server's storage device being executed by a processor, and the specified processes are realized in cooperation with other hardware. Note that functions equivalent to those configured by software can also be realized by hardware such as an FPGA (Field Programmable Gate Array) or an ASIC (Application Specific Integrated Circuit).

[0117] When the central server 240 performs control, it has multiple wind power generators 1 under its control, so the following processing is performed for each wind power generator. This processing is basically a repeated process, and the start timing is set by a timer or the like, for example, starting at midnight every day (S601). Alternatively, the central server 240 can perform the processing at any timing in response to an operator's instruction.

[0118] In step S602, the central server 240 checks when the lubricant needs to be replaced. The initial value of the replacement time can be calculated based on the physical properties, for example, by using the Arrhenius reaction rate, assuming that the lubricant is operating at its design temperature, and the remaining life can be set as an initial value. This replacement time can be updated later in step S610 based on actual measurement data.

[0119] If it is time to change the lubricating oil, the lubricating oil is changed in step S603. Since changing the lubricating oil is normally performed by a worker, the central server 240 displays and notifies the worker when and what to change.

[0120] If it is not time to change the lubricant, in step S604, the central server 240 performs a diagnosis using sensor data. Sensor data can include the near-infrared absorption spectrum of the lubricant obtained by an optical sensor, as well as oil temperature, oil pressure, and particle concentration in the lubricant, which can be measured using conventional technology. The data collected by the optical sensor 304 is sent to the central server 240, and the central server evaluates the physical properties of the lubricant by, for example, comparing the oil temperature, oil pressure, and particle concentration in the lubricant obtained from the sensor with predetermined thresholds.

[0121] If the diagnosis result in step S605 is abnormal, the lubricating oil is changed in step S603. If no abnormality is found, step S606 is carried out.

[0122] In S605, machine learning using correlations of near-infrared absorption spectrum data determines that there is an abnormal viscosity when the viscosity estimated from the near-infrared absorption spectrum measured by the optical sensor exceeds a predetermined threshold.

[0123] In step S606, the near-infrared absorption spectrum and the like are input to the central server 240, and the data is stored in chronological order.

[0124] From the perspective of preventive and planned maintenance of wind turbines, it is desirable to perform predictive diagnosis of lubricant deterioration based on changes in the viscosity or acid value of the lubricant before determining that an abnormality exists.

[0125] The replacement timing estimation result obtained in step S607 can be displayed as a lubricant diagnosis result (step S608).

[0126] 18 shows an example of the display results of process S610. It is recommended that this lubricant be replaced when its viscosity exceeds 200 cP. Based on the near-infrared absorption spectrum data acquired by the optical sensor, it is predicted that the viscosity will reach 200 in 50 months, so the new replacement time can be set to a time before that (for example, two weeks before). Process S610 completes one cycle of processing, and in process S602 of the next cycle, a determination process is performed in accordance with the new replacement time.

[0127] According to the above-described embodiment, by installing an optical sensor in a machine that uses lubricating oil, the physical properties of the lubricating oil, such as viscosity, acid number, and base number, can be quantified by the sensor without contacting the lubricating oil.

[0128] According to the above embodiment, efficient oil maintenance management can be realized, which reduces energy consumption, reduces carbon emissions, prevents global warming, and contributes to the realization of a sustainable society.

[0129] Diagnostic system 1200, estimation model generation unit 1250, oil property estimation unit 1260, actual measurement data collection process S1310, property estimation model generation process S1320, oil property estimation process S1330

Claims

1. A method for generating an estimation model of oil physical properties, using an information processing system equipped with an input device, an output device, a processing device, and a storage device, wherein the information processing system is equipped with an estimation model generation unit that generates an estimation model that can be subjected to machine learning, and an absorption spectrum obtained by transmitting light with at least a portion of wavelengths between 800 nm and 2500 nm through an oil containing a base oil and additives, and a value reflecting the physical property values ​​of the oil, are input to the estimation model generation unit, and an estimation model is generated by machine learning using values ​​based on the absorption spectrum as explanatory variables and values ​​reflecting the physical property values ​​of the oil as objective variables.

2. The method for generating an estimation model of oil properties according to claim 1, wherein the estimation model generation unit generates the estimation model using multivariate analysis, which is a type of machine learning.

3. The method for generating an estimation model of oil physical properties according to claim 2, wherein the estimation model generation unit performs multivariate analysis using PLS regression analysis.

4. The method for generating an estimation model of oil physical properties according to claim 1, wherein the wavelength of the light is in the range of 1550 nm to 1950 nm.

5. The method for generating an estimation model of oil physical properties according to claim 1, wherein the physical property values ​​of the oil are one or more physical property values ​​selected from viscosity, acid number, and base number.

6. The method for generating an estimation model of oil properties according to claim 1, wherein the explanatory variables are values ​​obtained by differentiating the absorption spectrum one or more times.

7. A method for generating an estimation model of oil properties as described in claim 1, wherein the estimation model generation unit verifies the estimation model by cross-validation.

8. A method for diagnosing oil properties using a diagnostic system consisting of an information processing device that implements the estimation model described in claim 1, comprising: inputting into said diagnostic system, as the diagnostic object absorption spectrum, an absorption spectrum obtained by transmitting light of at least a portion of wavelengths between 800 nm and 2500 nm through an oil to be diagnosed of the same type as the oil containing the additive, inputting a value based on said diagnostic object absorption spectrum into said estimation model, and obtaining from said estimation model a value that reflects the physical property value of said diagnostic object oil.

9. The method for diagnosing oil properties according to claim 8, wherein the value obtained by differentiating the absorption spectrum to be diagnosed one or more times is input to the estimation model.

10. The method for diagnosing oil properties according to claim 8, wherein the oil property values ​​are one or more properties selected from the group consisting of viscosity, acid number, and base number.

11. A method for diagnosing oil properties as described in claim 10, wherein, when the oil property value is viscosity, the value reflecting the oil property value obtained from the estimation model is corrected using the temperature of the oil at the time the absorption spectrum of the oil to be diagnosed is measured.

12. A method for diagnosing oil properties as described in claim 11, wherein, when the physical property value of the oil is viscosity, the viscosity at a reference temperature is calculated using the temperature of the oil to be diagnosed at the time the absorption spectrum of the oil to be diagnosed is measured, the viscosity obtained from the estimation model, and previously calculated temperature characteristic information of viscosity.

13. A method for diagnosing oil properties as described in claim 12, wherein, when the physical property value of the oil is viscosity, the dynamic viscosity at the reference temperature is calculated using the temperature of the oil to be diagnosed at the time the absorption spectrum of the oil to be diagnosed is measured, the viscosity at the reference temperature, and previously calculated temperature characteristic information of density.

14. An oil property diagnostic system comprising an information processing device that implements the estimation model described in claim 1.

15. The oil property diagnostic system according to claim 14, wherein the oil property value is one or more property values ​​selected from viscosity, acid number, and base number.

Images