Hidden info uncovered for machinery health
A data mining approach that correlates chemical data of lubricants with machinery performance in terms of lifetime and efficiency.
Every day, real-time maintenance, and industry 4.0 produce huge data from numerous analytical and sensorics tools in all industrial sectors. However, this information is highly specific, e.g., single parameters are reported for lubricants and lack from correlation with machinery performance and health. It is a challenge to produce a full picture of all data and to filter and correctly interpret relevant information for condition monitoring and maintenance.
Multivariate statistics has entered the field of tribology to support conventional data evaluation as well as to facilitate the detection of additional – “hidden” – information. Such tools are highly valuable as lubricants are composed of a wide range of chemical components. During application lubricants suffer from additive consumption, contamination, and formation of degradation products. Altogether, these processes determine the usability of a lubricant. Depending on the application, specific lubricant parameters play a major role regarding energy efficiency expressed as friction and remaining lifetime determined by residual additive amounts. For example, the availability of anti-wear additives keeps machine wear at a low level and hence contribute to machinery health.
For this purpose, a data mining approach was developed to extract important information and associations between lubricant chemistry and machinery performance. Chemical and tribometrical characterization of fresh, used and artificially altered engine oils were evaluated using multivariate statistical methods established in the field of chemometrics.
FTIR (Fourier-Transform Infrared Spectroscopy) spectra were used to compare the condition of engine oils that were artificially altered in the laboratory with those of used engine oils collected from vehicles. This preliminary work revealed similarities and differences in the chemical composition of artificially altered and used engine oils, which were necessary to automatically find associations between the engine oils’ compositions and their tribological performance when applied in machinery. Next, empirical quantitative models were built from these associations to predict lubrication and machinery performance from oil condition.
Impact and effects
Data mining approaches such as the presented one are used to automatically reveal associations and immediately provide reliable predictions based on lubricant data, here measured by conventional and easy-to-obtain lubricant analytics and sensors. Large data sets in the laboratory correlation models are created to link lubricant condition with lubrication and machinery performance in short time and at low costs. Coherences “hidden” for the “naked eye” on the first sight are made accessible and exploited for prediction of relevant behavior and consequently machinery health. When using these additional tools in industrial applications, predictive maintenance can be established to support sustainability and resourcesaving operation of lubricated machinery.
This data mining approach makes trends visible on a quantitative level and enables reliable information, for example, forecast about remaining lifetime and performance of different lubricant formulations. Moreover, it simplifies assessment and benchmarking on a quasi-standardized level without the need of highly complex methods for lubricant analysis.
Project coordination (Story)
Dr. Charlotte Besser
AC2T research GmbH