No abstract

No abstract

This document is the author's post-print version, incorporating any revisions agreed during the peer-review process. Some differences between the published version and this version may remain and you are advised to consult the published version if you wish to cite from it.

No abstract

To counteract thermal error due to material deformation in multi-axis heavy-duty CNC machining applications, this paper presents an approach based on Autoencoder-Augmented Long Short Term Memory deep learning models. The model predicts the z-axis offset of machine tools, in an idle state, with up to 96% accuracy, from ambient temperature sensor readings. By enabling a Long Short-Term Memory (LSTM) algorithm to learn time dependencies across input features and past output values with a fifth-order lag, a sequence-to-sequence regression problem is formulated. The model objective is to predict the z-axis thermal offset as an output from temperature sensor inputs. To reduce input data redundancy, the algorithm also leverages an Autoencoder to learn a reduced dimension feature representation for the input data. The reduction in input dimensionality achieves a considerable reduction in computation effort, measured in units of time, without compromising the prediction accuracy. The performance of the proposed technique is evaluated on a subset of the recorded data, and the model is benchmarked against several techniques from the literature. The results and qualitative evaluation demonstrate the superiority of the proposed approach at generating accurate predictions of the z-axis error of the system and, consequently, reducing the control error.