This project addresses the need for accurate friction process models during all operating conditions of a machine for friction- based diagnostics and motion control. Slow or sudden changes in friction characteristics due to wear, temperature, humidity, lubricant conditions, and loads can lead to undetected changes in the parameters of a governing friction process model. For model- based diagnostics, accurate friction process models are required to detect faults that may lead to machine failure. For motion control, friction process model uncertainty must be minimized to maximize performance. This research develops estimation methods for mechanical systems with nonstationary sliding or rolling friction to detect and compensate for changes in friction process models. To accomplish this, a synergistic combination of non- model-based and model-based estimation is developed to separate friction force observation from friction process model identification. For friction force observation, nonlinear estimators that do not require a structured friction model or direct measurement of friction force will be developed. Friction observers will use models of the physical system (apart from the friction model) and measured motion to estimate friction force through state extended filtering. Friction force observation will supply inputs for model-based, multiple-model estimation to identify a physically relevant friction model using Bayesian selection. Experiments combining estimation and control will focus on maximizing accuracy of high-speed, repetitive, motion control for diverse applications such as semiconductor manufacturing, positioning hard disk read/write heads, and machine tool position/velocity control. Use of the estimation methods to enhance friction based diagnostics will be demonstrated by simulation in the context of model-based condition monitoring of bearings.