Data for: Simulator-assisted deep learning framework for anomaly detection in continuous solvent extraction for reliable metal recovery in lithium-ion battery recycling
Description
This dataset provides simulator-generated pH trajectories used for developing and evaluating deep-learning-based anomaly detection models in a continuous solvent extraction process for lithium-ion battery recycling. The dataset consists of ten normal-operation trajectories generated under volumetric-flow perturbations and twenty fault-operation trajectories generated under predefined process disturbances. The fault trajectories include step and ramp disturbances introduced at specified process sections after 700 min of normal operation. Each trajectory contains time-series pH data from seven settler outlets sampled at 12 sec intervals. The provided data correspond to simulator outputs before artificial pH measurement noise was added. Measurement-noise injection, normalization, model training, and evaluation were performed in Python and are described in the associated code repository.
Files
Steps to reproduce
The released Excel workbooks contain simulator-generated base trajectories. The normal-operation trajectories are provided in `Normal_trajectories.xlsx`, and the fault-operation trajectories are provided in `Fault_trajectories.xlsx`. Variable descriptions, trajectory indexing, data split information, fault-scenario mapping, and preprocessing notes are provided in `metadata.xlsx`. The simulator outputs were recorded at 12 s intervals. In the associated anomaly-detection workflow, the trajectories were downsampled to 1 min intervals during Python preprocessing. Gaussian pH measurement noise with a standard deviation of 0.03 pH units was then added to the pH variables before normalization. To reproduce the data-processing workflow, users should first load the trajectory sheets listed in `metadata.xlsx`, downsample the simulator outputs to 1 min intervals, generate Gaussian pH-noise realizations for the normal-operation trajectories, and apply the train/validation/test split described in the `train_validation_test_split` sheet. The associated code repository provides scripts for downsampling, pH-noise injection, preprocessing, normalization, model training, model inference, anomaly-score calculation, thresholding, and performance evaluation.
Institutions
- Chonnam National UniversityGwangju, Gwangju