Falling down can cause moderate to severe injuries, increasing the risk of death among elderly. For this reason there is a substantial growth of Ambient Assisted Living technologies, including smart environments, in order to support elderly and fragile people in potentially dangerous situations. The paper describes an expert system based on a wireless wearable low-cost accelerometer able to automatically detect falls, generalizing the detection of critical events in several practical conditions. The algorithmic scheme appears invariant to age, weight, height of people and relative positioning area (even in the upper part of the waist), resulting compliant with many commercial wearable devices. Experimental results show high generalization properties and better performances than well-known threshold-based approaches.