This is the complementary repository to the paper "Learning Process Steps as Dynamical Systems for a Sub-Symbolic Approach of Process Planning in Cyber-Physical Production Systems" accepted at the HYDRA Workshop at the European Conference on Artificial Intelligence (ECAI) 2023.
The repo contains an implementation of the Process Step Representation network architecture (peppr), and all supplementary material and datasets to reproduce the results of the paper.
Approaches in AI planning for Cyber-Physical Production Systems (CPPS) are mainly symbolic and depend on comprehensive formalizations of system domains and planning problems. Handcrafting such formalizations requires detailed knowledge of the formalization language, of the CPPS, and is overall considered difficult, tedious, and error-prone. Within this paper, we suggest a sub-symbolic approach for solving planning problems in CPPS. Our approach relies on neural networks that learn the dynamical behavior of individual process steps from global time-series observations of the CPPS and are embedded in a superordinate network architecture. In this context, we present the peppr architecture, a novel neural network architecture, which can learn the behavior of individual or multiple dynamical systems from global time-series observations. We evaluate peppr on real datasets from physical and biochemical CPPS, as well as artificial datasets from electrical and mathematical domains. Our model outperforms baseline models like multilayer perceptrons and variational autoencoders and can be considered as a first step towards a sub-symbolic approach for planning in CPPS.
You can access the paper here
Python and venv requirements cf. peppr.yml.
To install the venv with Anaconda: (1) change the path in peppr.yml to the directory you want to install the venv. (2) Open the file location in your terminal and type mamba env create -f peppr.yml.
We evaluated our model on four different datatset. The datasets can be accessed under the following links.
| dataset | description | link |
|---|---|---|
| polynom | Sinus function modulated by a polynomial functions of different degrees | https://github.com/j-ehrhardt/basic-datagenerator |
| ode | Data from a PT-1, PT-2, and PT-3 unit | https://github.com/j-ehrhardt/ode-ml-datasets |
| BeRfiPl | A benchmark about the physical behavior of Cyber-Physical Process Systems. | https://github.com/j-ehrhardt/benchmark-for-diagnosis-reconf-planning |
| IndPenSim | A dataset of a Penecillin fermentation process | http://www.industrialpenicillinsimulation.com/ |
All necessary code to run and evaluate the models is in the ./model directory.
To train the models, save the datasets into the equivalent directories in the ./data/dataset-name directory.
Modify the experiments.json file in the ./model directory, and run the main.py.
You can create a gridsearch by defining the parameter ranges of the experimental_setup.py file, creating a new experiments.json file that includes all experiments of the gridsearch.
The data_module.py file contains all necessary dataimport and dataloaders.
The eval_module.py file contains the evaluation metrics.
The train_module.py file contains the training procedures for all network architectures.
The module.py file contains the network architecture components, the networks are defined in the nets.py file.
The ultils.py and vis_utils.py files contain utility and visualization functions.
When using the code from this paper, please cite:
@ARTICLE{Ehrhardt2024-ke,
title = "Learning process steps as dynamical systems for a sub-symbolic
approach of process planning in {Cyber-Physical} Production
Systems",
author = "Ehrhardt, Jonas and Heesch, Ren{\'e} and Niggemann, Oliver",
journal = "Commun. Comput. Inf. Sci.",
publisher = "Springer Nature Switzerland",
pages = "332--345",
year = 2024,
}
Licensed under MIT license - cf. LICENSE