Time prediction for flexible manufacturing
Functions for training and evaluation of a stacked LSTM model to predict the time to produce a given item.
The code provides functions for training and evaluation of a stacked LSTM model with spatial dropout. It predicts the duration of production per item for a given production plan. The products have different configurations (in this example we have three different colors). In the prepocessing the function can add padding to the sequence of products such that the LSTM trains well even if the starting time of the items is irregular.
Install & Run: To generate input data with the simulation call 'python ./beltsimulation.py output_filename.csv'. The jupyter notebook 'train_and_evaluate.ipynb' should be self-documenting.
Additional information: A simulation for the input data is included in the code. It serves as a basis for an uncertainty estimation (not covered in this code).
This code is the result of the collaboration of the AI4EU Task 6.4 (AI4Industry) and Task 7.2 (VerifiableAI).