With the rapid advancement of large language
models (LLMs), the integration of knowledge-driven and gen
erative AI into various industries has emerged as a prominent
research focus. In the field of tribology, several longstanding
challenges persist. One major issue is the variability of wear
conditions across different friction pairs, necessitating separate
model training for each working condition. This approach is
often constrained by limited training data, resulting in reduced
usability and the common problem of black-box models lacking
interpretability. Recent research suggests that LLMs possess
certain reasoning and abstraction capabilities, enabling them
to incorporate prior knowledge into their outputs for more
explainable results. Motivated by this, we constructed a fine
tuning dataset specifically for friction and wear, and used
it to train a lightweight large model for wear prediction.
Experimental results demonstrate that our model outperforms
mainstream medium-parameter LLMs deployed on edge de
vices.