Triboinformatics or Intelligent Tribology

The subjects of tribology, which is defined as the study of the interaction of solid surfaces in their relative motion, are friction, wear, lubrication, surface properties, and adhesion. Since the term tribology was introduced by Peter H. Jost in 1966, the field has developed and yielded such new areas of study as nanotribology, biotribology, ecotribology (sometimes called “Green Tribology”), biomimetic tribology, and many others. However, one of the main challenges in the tribological studies is that while there is an abundance of data about the frictional, wear, and surface properties of various materials, systems, and engineering components, this interdisciplinary area is highly empirical. While many attempts have been made to formulate various laws or rules in tribology, the study of friction and surface properties often lacks derivation from physical or chemical first principles and, consequently, tribology remains data-driven inductive science.

This is not a new problem and it is not specific to tribology. Many areas of scientific research have been developing as collections of empirical data. The philosophers of science already in the 19th century have distinguished between nomothetic and idiographic scientific disciplines. The former are concerned with the study and discovery of general scientific laws, while the latter deal with the study and discovery of particular scientific facts and processes. In this classification, tribology would perhaps be designated as an idiographic discipline, because its primary method is empirical and its main concern is the study of particular facts related to friction, wear, lubrication, surface characterization, and adhesion.

Since the onset of the 21st century, the relationship between the scientific facts and scientific laws has changed. This is because of the emergence of powerful computational technologies and methods, which have led to the new paradigm of data analysis often dubbed the Big Data. The Big Data paradigm uses inductive statistics to infer laws, regressions, nonlinear relationships, and causal effects from large sets of data with low information density. This allows to reveal relationships and dependencies or to perform predictions of outcomes and behaviors. Big Data employs such techniques as machine learning and artificial intelligence. The correlations found by the Big Data analysis do not necessarily reflect causation, which is a major departure from the classical scientific methodology developed since the time of the Scientific Revolution of the 17th century, which assumed that scientific laws are behind causation.

One particular area of interest is the multidimensional Big Data, which can be represented as data cubes or “tensors” (a multidimensional generalization of the spreadsheet table). Specific computational methods have been developed to retrieve correlations in large data sets. For example, the topological data analysis is used to reduce the dimension of high-dimensional datasets using persistent homology and to find low-dimensional structures with reproducible topology (the Betti numbers). An example in mechanics would be the trajectory patterns called Lagrange Coherent Structures (LCSs), a computational fluid mechanics concept which has emerged and became popular in the past 20 years. The LCSs are somewhat similar, but not identical to the classical dynamic concept of the invariant manifolds.

Driven by emerging technology, Big Data methods in engineering are closely related to novel concepts such as the Internet of Things, embedded real-time sensors and Digital Twins. Personalized medicine and diagnostics using the Big Data methods is expected to make an impact on the health care industry.

The Big Data approach is the mainstream in various data-rich natural sciences. Thus in microbiology, new disciplines often informally referred to as “omics” have emerged. They include, among others, genomics (the study of the genome or the set of genes), proteomics (the study of proteins and relations between them), connectomics (the study of networks of neurons), immunomics (the study of networks of immune cells, antigens, and antibodies), and metabolomics (the study of metabolism). The aim of “omics” disciplines is the collective characterization and quantification of pools of biological data that translate into the structure, function, and dynamics of organisms.

Similar tendencies are found in social studies, where data mining (such as observations of individual behaviors, for instance, at the internet or mobile device usage) is used to explain and predict human behavior for various purposes ranging from making advertisement more effective to fighting terrorism or tracing individuals during the Covid-19 pandemic. Moreover, even in humanities, the current trend is the so-called Digital Humanities, which often involve Big Data approaches to cultural artifacts.

One particularly interesting new data-driven area is network science, where a number of amazing discoveries have been recently made including the clustering, small-world, and scale-free networks (“Clustering and self-organization in small-scale natural and artificial systems”; Scaling in Colloidal and Biological Networks). Network approaches have already been used in materials science and colloidal science to characterize granular media, colloidal crystals, and droplet clusters. The biomimetic Artificial Neural Networks (ANN), which is a set of algorithms of machine learning loosely based on the functioning of neurons in the human brain, constitute an example of a method to establish correlation trends between loosely defined data. The ANN models improve their ability to estimate the output without specific programming on the task but by comparing their predictions with new data. Similarly to human beings who learn associations by way of numerous examples, ANN use examples and training to learn. They store the knowledge of the input-output relationships acquired in the learning process in the synaptic weights of the inter-nodal connections.

All these developments suggest that it may be about time to apply the Big Data concepts for tribology and surface science, which are data-rich areas. This development could potentially lead to a new sub-area of tribology, which has already been called by some practitioners as “Intelligent Tribology” or “Tribo-informatics”.

Friction is often viewed by physicists as a non-linear phenomenon inherently related to the reduction of the degrees of freedom. The large number of degrees of freedom, which characterize molecules or microasperities, is reduced to the minimalistic models involving a small number of most relevant degrees of freedom characterizing motion and sometimes internal state variables]. The mimimalistic models with a small number of degrees of freedom reflect the dynamic behavior of a frictional system. However, in the case of friction there is no simple way of reducing a molecular-scale model to the minimalistic model. The traditional statistical mechanics and thermodynamics methods do not always work well for this task. For example, there are difficulties in deriving the dry friction laws from the Second Law of thermodynamics. This is the area, where the topological data analysis and other methods such as neural networks could provide new insights.

Our new paper:

Amir Kordijazi, Hathibelagal M Roshan, Arushi Dhingra, Marco Povolo, Pradeep K Rohatgi, Michael Nosonovsky, “ Machine-learning Methods to Predict Wetting Properties of Iron-Based Composites” Surface Innovations 1-8