New tools are needed to reach high goals for uptime and availability in industrial processes. Early warning of developing faults is one part of the strategy to reach these goals. A single method rarely meets all requirements, but combining methods and techniques in a hybrid system offers advantages and can overcome limitations in the individual approaches. Methods considered are physical models, artificial neural networks, and case-based reasoning. The paper discusses the pros and cons, strengths and weaknesses of the three methods and three combinations of hybrid solutions in order to assist in select a suitable combination for a specific early warning challenge ahead.