Target position joint angles of industrial robot armvmodel estimating by using constructive neural network and application of controlled trajectory

The coordinate of the end function of the industrial robot arm at a given point in cartesian coordinates and the operation of the robot arm's joint angles and the working angles of the robot are known as kinematic processes. Inverse kinematic solutions, especially in multi-jointed robots, present difficulties due to the mathematical operations involved. In this direction, it is aimed to estimate the final joint angle values of the robot arm using the constructive neural network (CoANN) without inverse kinematic analysis. Thus, problems / difficulties arising from advanced and inverse kinematic solutions have been eliminated. In this study, industrial robot arm model consisting of four joints and one holder was used. CoANN models were compared to traditional artificial neural network (ANN). In the traditional ANN model contains a 3-layer structure and 50 cell per layer. CoANN model works prefer pyramid structure. The training of the CoANN started with a cell, and the training grew to reach the target training regression value and completed the training in 59 layers. 5000 training data is applied in ANN trainings. The tests in the study are for the given fixed points of the robot arm and for the instant coordination of the given trajectories. Both ANN models were able to compute a high degree of joint angles for the given end functional coordinates except for the training set. Thus, inverse kinematic equations were not used and problems experienced in inverse kinematic solutions were not experienced.