Elsevier

Talanta

Volume 219, 1 November 2020, 121312
Talanta

Artificial Intelligence and fourier-transform infrared spectroscopy for evaluating water-mediated degradation of lubricant oils

https://doi.org/10.1016/j.talanta.2020.121312Get rights and content

Highlights

  • Simple and fast detection of lubricant oil aging.

  • Simple and fast detection of water presence in lubricant oils.

  • Artificial Intelligence applied to FTIR data for detecting damage in lubricant oils.

Abstract

The presence of water in lubricant oils is a parameter related to the lubricant deterioration, which can be indicative of a serious loss of tribological efficiency and, therefore, an increase in maintenance costs. Likewise, controlling the aging of the lubricant oil is a keynote issue to prevent damage on the lubricated surfaces (e.g. engine pieces). The combination of Attenuated Total Reflectance (ATR) techniques with Fourier-Transform Infrared Spectrometry (FTIR) result in an easy, simple, fast and non-destructive way for obtaining accurate information about the actual situation of a lubricant oil. The analysis of this ATR-FTIR information using Artificial Neural Networks (ANN) as well as Linear Discriminant Analysis (LDA) results in the proper classification of lubricant oils regarding the presence/absence of water, age and viscosity. The methodology proposed in this work describes procedures for identifying the deterioration degree of oils with as high as 100% success (aging week) or 97.7% (for viscosity and water presence).

Introduction

Friction causes important losses of money. It is estimated that friction and related phenomena in industrialized countries cost up to €450 billion annually, despite the wide use of lubricants [1]. According to Webster dictionary, a lubricant is “a substance (such as grease) capable of reducing friction, heat, and wear when introduced as a film between solid surfaces” [2]. The idea behind a lubricant, therefore, is the creation of a film between the sliding parts of an engine (or other moving device), filling the space between the surfaces and keeping them apart. A lubricant material has to achieve some objectives: it has to improve machine-performance by reducing mechanical energy, withstand high temperatures, maintain its viscosity and clean up the impurities that are generated during the sliding of mechanical parts.

Basically, a lubricant consists in a base oil and a series of additives, which depend on the use the lubricant is designed for. Currently, three main types of base oils that differ from each other by their origin can be considered: mineral oils, synthetic oils and semi-synthetic oils. i) Mineral oils that proceed from the fractional distillation of crude oil and are elaborated through multiple processes that generate adequate products to form base oil. ii) Synthetic lubricants (such as poly-α-olefins) can be manufactured using chemically modified petroleum components or synthetic esters produced from other raw materials. Synthetic oils are used in those applications with specific equipment demands that conventional mineral oil do not fulfil and when economic benefits are offered [3]. iii) The composition of a semi-synthetic lubricating oil comprises a major portion of a synthetic lubricant or a mixture of a synthetic lubricant plus a hydrocarbon mineral base oil, and a minor portion of various additive components [4]. Mineral oils are one of the main liquid lubricants used in industrial machinery and in automotive industry with an annual expenditure of approximately 9000 million liters [5].

Corrosion and degradation behavior of oil lubricants is critically related to water content. Water contamination in lubricant oils may coexist in a combination of forms: free water, dissolved in oil and emulsified. Free water is that which exists in excess of its equilibrium concentration in solution. Dissolved water is simply water in solution and its concentration in oil depends on the oil hygroscopic properties, temperature and humidity. Water out of the limits imposed by these conditions is free water. If the free water content increases the lubricant oil becomes saturated with enough water molecules to a point that water is suspended in micro/nanoscopic droplets giving rise to an emulsion, often undesirable. Free water and emulsified water are the most damaging of all water-lubricant mixtures. So, free water may settle on machinery surfaces displacing any protective lubricant film, thus contributing to the corrosion of the surface [6]. On the other hand, emulsified lubricants exhibit reduced load carrying capacity due to a change in the compressibility of the oil. Besides, the emulsified water has a tendency to trap dirt and particulates creating a sludge that may grind the contacting surfaces creating more particulates. The consequent lubricant failure is then followed by permanent damage to the operating surfaces [7,8]. Besides, water contamination promotes chemical and physical changes in the lubricant oil properties. In fact, water plays a key role in the increase of lubricant aging rate, depletion of additives and destruction of base oils causing acid formation [9,10]. In some instances, there is a need to replace the lubricant oil or use demulsifiers. For those applications where the lubricant must be regularly replaced, there is a direct relationship between lubricant deterioration and consumption, along with the environmental impact of lubricant disposal [11]. Consequently, if an adequate control of the water content and lubricant stability is made before it reaches the point of causing corrosion or losing effectiveness, a better lubricant economy, a reduction in friction losses and hence decreased environmental hazards are possible [12,13].

The development of effective and efficient methods to assess the aging of lubricant is important to the lubrication industry. It is desirable that the methods do not require sample preparation and produce rapid results and information on multiple parameters simultaneously. Among the techniques that provide functional-group information Fourier Transform Infrared Spectroscopy (FTIR) allows to ascertain their presence in a sample, as each group absorb in defined wavelength regions. So, chemical changes taking place in lubricant oils over time are associated with the formation of new or loss of particular functional groups due lubricant oil aging. FTIR is rapid and non-destructive technique. However, FTIR is not sufficient to provide a plausible differentiation criterion if chemical mapping is applied to large data sets that include spectra from many lubricant oils. In the last ten years, research has been carried out to apply FTIR combined with chemometrics to address this shortcoming by analyzing large amounts of spectral information. The viscosity index (VI) and the base number (BN) of motor oils were successfully determined using the FTIR analyses along with the PLS-1 calibration. Gracia et al. applied FTIR spectroscopy combined to Principal Component Analysis (PCA) in order to extract chemical information during oil oxidation process in the presence of iron as a catalyst. Results demonstrated that the presence of iron did not lead to significant change in the global chemical composition as revealed by the first principal component. However, the second principal component indicated clearly that the initial formation of alcohols and esters was favored by the presence of iron. The influence of iron was highlighted by PCA analysis of on-line FTIR data [14]. In other study, Nguele et al. [15] studied the depletion of additives in lubricating oils due to their degradation through a series of chemical reactions which resulted in loss of their primary functions. To tackle the subject, authors used FTIR-ATR spectral data combined with a curve fitting technique and mathematical models, that describe the behavior of additives within the engine. Results showed that depletion of additives followed an exponential regression rather than polynomial and that the chemical breakpoint (the initiation of deterioration of additives) depended on the composition of the base stock. The breakpoint was found to be two times higher in a fully synthetic model lubricating oil than a semi synthetic model one. In 2018, Hossain et al. [16], applied FTIR in connection with Artificial Neural Network (ANN), Principal Component Regression (PCR) and Partial Least-Square Regression (PLSR) for determination of VI of motor oils. Results showed that among the calibration techniques studied, PLSR provided the best prediction results with Savitzky-Golay smoothed FTIR spectral data, the method requiring shorter turnaround times and lower expenses than conventional approaches.

In this work, we compared the performance of non-linear (ANN) and linear calibration techniques (Linear Discrimination Analysis, LDA) for prediction of the aging degree of lubricant oils from Attenuated Total Reflectance-Fourier-Transform Infrared Spectroscopy (ATR-FTIR) spectra. The model was created for four important lubricant oils properties: water content, viscosity, oxidation and time of the experiment. ANN demonstrated to be a good chance to identify base oils according to their viscosity, as well as to detect the presence/absence of water with a fast, simple and non-destructive ATR-FTIR measurements. Also, ANNs have an outstanding performance for detecting the aging time of the base oil. ANN classify without error samples in periods of one week and show a good trend to group samples in shorter periods of times. Results thrown by LDA were compatible and coherent with those obtained by ANN. In our knowledge, ATR-FTIR aided by ANN has never been used to predict the contamination by water and the aging time of lubricant base oils.

Section snippets

Oils samples and instrumentation

Base oils with different viscosities were kindly provided by REPSOL S.A. A Varian 670-IR spectrometer equipped with a DLaTGS detector and a diamond-based Golden Gate ATR device, with an internal reflection (crystal area 1 × 1 mm), was employed for all the measurements. The spectrometer was completely software-controlled by the Varian Resolutions Pro software provided by Varian Inc. Mathematical data processing and calculations were performed with MatLab®. Linear Discriminant Analysis and

Results and discussion

Mid-IR spectra provide information about the functional groups present in the different lubricant oils. The assignation of the most intense bands in the Mid-IR spectra of the base oils to the different functional groups is listed in Table 1. These data demonstrate the aliphatic nature of the hydrocarbon chains in the base oils (bands at 2952, 2920, 2871, 2852, 1458, 1378, 723 cm-1), as well as the presence of water (3410 cm-1), aldehyde groups (2729 cm-1), double bonds Cdouble bondC (1610, 1305, 1157,

Conclusions

Artificial Neural Networks are a good chance to identify base oils according to their physical and chemical propierties, as well as to detect the presence/absence of water. This is an interesting tool to detect the oil contamination by water with a fast, simple and non-destructive FTIR measurements. Also, ANNs have an outstanding performance for detecting the aging time of the oils. ANN classified without error samples in periods of one week and showed a good trend to group samples in shorter

Funding

This work was supported by the Spanish Ministry for Economy and Competitiveness (Ministerio de Economía y Competitividad) and the European Regional Development Fund [MINECO/FEDER, projects # MAT2015-66747-R and # RTI2018-099756-B-I00]; and the Foundation for the Promotion in Asturias of the Applied Scientific Research and Technology [FICYT project # GRUPIN14-023].

CRediT authorship contribution statement

Christian Chimeno-Trinchet: Data curation, Formal analysis, Software. Clarissa Murru: Data curation, Formal analysis, Validation. Marta Elena Díaz-García: Investigation, Supervision, Writing - review & editing. Alfonso Fernández-González: Writing - original draft, Methodology, Data curation. Rosana Badía-Laíño: Funding acquisition, Methodology.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Acknowledgements

Authors are gratefully acknowledged to Ministerio de Economía y Competitividad, the European Community and the Foundation for the Promotion in Asturias of the Applied Scientific Research and Technology) for the economic resources. We want also to thank the (Photo)electronic and Vibrational Spectroscopy Unit from the Scientific and Technological Resources of the Universidad de Oviedo for the FTIR measurements.

References (28)

  • J.D. Olden et al.

    An accurate comparison of methods for quantifying variable importance in artificial neural networks using simulated data

    Ecol. Model.

    (2004)
  • K. Holmberg et al.

    Influence of tribology on global energy consumption, costs and emissions

    Friction

    (2017)
  • Lubricant in merriam-webster

  • 32 United States Patent (19) Hammond et al. 54 SEMI-SYNTHETICLUBRICATING OIL COMPOSITION Kenneth G. Hammond,...
  • Cited by (18)

    • Electrical compatibility of transmission fluids in electric vehicles

      2022, Tribology International
      Citation Excerpt :

      The evolution of electrical conductivity during the ‘AO consumption phase’ and the ‘oxidation phase’ should be determined to correlate the oxidation state with the conductivity values at longer term. On the other hand, Chimeno-Trinchet et al. [26] evaluated the mid-infrared bands which seems to be correlated to oil aging using artificial neural networks, identifying that absorptions at 2920 cm−1, 889 cm−1 and 812 cm−1 have the strongest discriminating ability towards the oil aging. However, in the present work the differences in the absorbance at these wavenumbers are nearly undetectable, with differences below 5% of absorbance, suggesting that the aging process followed different pathways.

    • Analyses of used engine oils via atomic spectroscopy – Influence of sample pre-treatment and machine learning for engine type classification and lifetime assessment

      2021, Talanta
      Citation Excerpt :

      Apart from that, supervised methods have previously been used to predict engine oil characteristics, e.g. contents of cheap engine oil in adulterated oils via partial least squares regression [32], thermophysical and physicochemical properties via support vector regression [34,35] or the total acid number of various engine oils using several machine learning algorithms [36,37]. Furthermore, artificial neural networks models have been employed to predict physicochemical properties and oil ageing from infrared spectroscopy [38]. Although different machine learning algorithms may be suitable for specific analytical questions, the best performance for big datasets can be achieved by model ensembles that combine many learner variations.

    • Computer-vision classification of corn seed varieties using deep convolutional neural network

      2021, Journal of Stored Products Research
      Citation Excerpt :

      This model uses the gradient steepest descent method to adjust neuron weights and minimize output error (Kuo et al., 2020). For example, Chimeno-Trinchet et al. (2020) used a scaled conjugate gradient backpropagation algorithm to train the network. The optimal number of hidden layers was 6.

    • Smart carbon dots as chemosensor for control of water contamination in organic media

      2021, Sensors and Actuators, B: Chemical
      Citation Excerpt :

      In the field of oil-based lubricants, moisture (water) contamination reduces its chemical stability and may cause many problems to machinery life such as corrosion, water etching (owing to by-products from lubricant degradation), vaporous cavitation (instantaneous vaporization and condensing implosion of water), hydrogen embrittlement (under extreme conditions water dissociates to release hydrogen atoms which then permeate the metal causing pinning and embrittlement of the metal) [7–9] among other effects. Control of water contamination in lubricants and in organic solvents can be carried out by different approaches such as FTIR-chemometrics, photoacoustic spectroscopy, electrochemical impedance spectroscopy, coulometric Karl Fischer titration, different types of sensors (capacity, conductance, permittivity, viscosity), fiber optic devices, etc [10–14]. The determination of moisture/water in organic solvents using luminescent dyes has been addressed in different fluorescent formats, including immobilization of fluorescent dyes on a surface (glass, polymer, etc), metal-organic frameworks, photo-induced electron transfer (PET), intramolecular charge transfer (ICT) or europium-dye nanospheres [15–18].

    • Effect of temperature on the heat treatment to recover green solvent from emulsion liquid membranes used in the extraction of Cr(VI)

      2020, Chemical Engineering and Processing - Process Intensification
      Citation Excerpt :

      Comparing all samples analyzed, fresh oil, recovered oils and W/O formulated emulsion slight differences are observed at around 800 and 3400 cm−1, being differences highly appreciated for the W1/O emulsion and less important in the case of oils recovered at different temperatures. Differences observed around 800 cm−1 can be attributed for differences on aromatic groups degradation which can be either oxidized by the presence of water or even evaporated due to the high temperature used, in the case of differences observed at around 3400 cm−1 were typically assigned to the presence of water in oil effluents [45]. Since the best DE values were obtained at temperatures of 70, 80 and 90 °C.

    View all citing articles on Scopus
    View full text