An Industrial Robot Data Analysis Method Based On PCA-ELM Algorithm

With the development of industrialization and modernization all over the world,industrial robots are more and more popular in many areas.We have witnessed a rapid development of robots in the past decade and today industrial robotic systems are key components of automation.The Robotic systems have many parameters that need to be tuned.The tuning task has traditionally been delegated to a human,and accomplished through a process of trial-and-error.Unfortunately,hand-tuned parameters tend to make a system brittle.Even if a particular set of parameters works well in one application,it may not easily transfer to other applications.In order for a robot to operate autonomously,it must be capable of interacting with its environment in an intelligent way.This implies that an autonomous robot must be able to collect data about the environment and then perform actions based on these data,that's the reason why a good data analysis method is the key point to the robotic system.An Improved PCA-ELM approach applied to error recovery in an industrial robot,which is used for palletizing,is proposed in the first part.To improve the existing error recovery method is the goal which is discussed,instead of the traditional expert system,the use of a learning algorithm to support the on-line error recovery is promoted.Due to the quick nonlinear processing capability and good generalization performance,we choose ELM as the classification algorithm.Besides,ELM has extremely strong learning ability,which means the algorithm is able to learn even with a minimal number of training examples in a very short time.However,ELM is not applied to the data sets with noise.In order to increase the forecast accuracy of ELM method,the PCA-ELM method is used to model.The feature of our new improved PCA-ELM approach bring a big advantage in real production since it would be impossible to accumulate a large number of training examples on site.Furthermore,it also greatly reduces the debugging time and the production cost.Therefore,the model has great value in practical use of industrial robot error recovery field.Firstly,the input and output signals and the measure to each situation were defined according to the on-site experiences.Secondly,the classification model was established by using the Improved PCA-ELM algorithm.Then,the validity of the approach was verified in a simulated environment.Finally,we discussed how the number of nodes of hidden layer influence the model according to the experiment results and how we put in use of model with different parameters.An‘corner path failure' detective model based on the improved PCA-ELM approach is proposed in the second part.‘Coner path failure' is a common fault in the industrial robotic system,however,there's no measure to detective this kind of fault during programming stage.The goal of this model is to help programmers find‘Coner path failure' during programming stage in order to avoid unnecessary mistakes during the debugging stage on site.The same method,improved PCA-ELM algorithm was used in the‘coner path failure'model.Firstly,the data were collected on sited.Secondly,the classification model was established.Then,the validity of the approach was verified in a simulated environment.