Data-driven multi-response optimization of locust bean pulp brake-pad composites using Taguchi L25 design, machine learning and SHAP explainability

Every time a car comes to a stop, a complex physical drama plays out between two surfaces. A brake pad, pressed against a spinning metal disc, must generate enough friction to slow the vehicle without wearing away too quickly or losing its grip when hot. For decades, these pads have relied on a mix of synthetic fibers, metals, and minerals to perform this delicate balancing act. Yet, as the world seeks to reduce its reliance on non-renewable resources, scientists are turning their attention to the natural world for answers. One promising avenue involves using agricultural waste—specifically, the fibrous pulp from the locust bean tree, a common sight across West Africa. The challenge lies in figuring out exactly how much of this plant material to mix with a binding resin, and under what conditions to bake the mixture, to create a brake pad that is both powerful and durable. It is a puzzle of ingredients and heat, where getting the recipe wrong could mean a pad that is too soft, too brittle, or simply ineffective.

In a recent study, researchers tackled this puzzle by treating the creation of a brake pad like a high-stakes cooking experiment, but with a heavy dose of modern data science. They focused on five key variables: the amount of locust bean pulp, the amount of phenolic resin binder, the size of the pulp particles, the pressure used to squeeze the mixture together, and the temperature at which it was baked. Instead of testing every possible combination of these five factors, which would have required thousands of trials, the team used a smart experimental design to run just twenty-five specific tests. This approach allowed them to map out how changing one ingredient or condition affected five different outcomes: how well the pad gripped the disc, how fast it wore down, how hard it was, how dense it became, and how much weight it could crush before breaking.

The researchers did not stop at simply recording the results of these twenty-five tests. They recognized that the relationship between the ingredients and the final performance is not a straight line; changing the amount of pulp might help one property while hurting another. To navigate this complexity, they fed their data into several advanced computer learning models. These digital tools acted as sophisticated predictors, learning from the twenty-five experiments to guess how the brake pads would behave under conditions they had not yet tested. The team compared different types of these learning algorithms to see which one was best at predicting each specific outcome. They found that no single computer model was perfect for everything; one type of algorithm was best at predicting friction, while another excelled at predicting hardness. By letting the best model handle each specific job, they created a highly accurate digital map of the material's behavior.

The analysis revealed a clear hierarchy in what mattered most. The amount of locust bean pulp turned out to be the single most influential factor, shaping almost every aspect of the brake pad's performance. Generally, adding more pulp made the pads harder, denser, and stronger, while also improving their grip and reducing how quickly they wore away. The temperature at which the pads were baked was the second most important factor, particularly for the mechanical strength of the final product. The study identified one specific combination from their twenty-five tests that performed the best in the real world. This winning formula contained thirty percent pulp, seventeen and a half percent resin, particles sized at twenty-five micrometers, compressed at forty megapascals, and baked at one hundred and eighty degrees Celsius. This specific batch achieved a friction coefficient of 0.476, a wear rate of 2.43, a hardness of 101, a density of 1257, and a compressive strength of 109.6.

However, the researchers were careful to distinguish between what they had physically made and what their computers suggested might be even better. While the best physical batch they created was the one described above, their digital models pointed to a slightly different, continuous range of ingredients that could theoretically yield an even more perfect result. This hypothetical "ideal" mix suggested a pulp content of roughly twenty-seven percent and a baking temperature near one hundred and ninety-four degrees. The computer predicted this mix would perform almost perfectly, with a desirability score of 0.9904. Yet, this remains a theoretical candidate. The study explicitly states that this digital optimum has not been physically built or tested. The researchers emphasize that before any car manufacturer could consider using such a formula, independent scientists would need to fabricate the material from scratch and verify that the computer's predictions hold true in the real world.

Ultimately, this work demonstrates a powerful new way to develop sustainable materials. By combining a small, carefully chosen set of physical experiments with explainable machine learning, the team was able to screen complex formulations quickly and identify the most promising directions. They proved that locust bean pulp is a viable reinforcement for brake pads and that its content is the dominant driver of performance. The study does not claim to have solved the problem of brake pad manufacturing, nor does it offer a final, ready-to-use product for the automotive industry. Instead, it provides a validated roadmap and a set of tools for future researchers. It shows that with the right data and the right computational lens, we can turn agricultural waste into high-performance engineering materials, provided we are willing to do the hard work of confirming the computer's guesses with real-world experiments.