Maintaining a good state of health is most challenging and necessary for all. In this regard, carrying out regular exercise with proper moves is important. Improper exercise moves often results is a strain, injury or no progress towards the goal. It is necessary to have a trained professional for proper guidance. Smart systems are in place, to help people in tracking their exercising activity, in the form of wearable bands or mobile apps. In this paper, an exercise assist tool is proposed to distinguish proper and improper exercise moves. Accelerometers were attached to upper limbs for data collections and an Artificial Neural Network (ANN) was used to classify them based on the features extracted. The proposed system was tested on five different volunteers for three different exercise movements. It was observed that the network was able to identify which axis the movement goes wrong. © 2017 The Authors. Published by Elsevier Ltd.