ANSR-DT

ANSR-DT is an adaptive neuro-symbolic framework for digital twins that combines temporal anomaly detection, symbolic reasoning, and reinforcement learning for interpretable monitoring and decision support. The repository contains the core ANSR-DT pipeline, symbolic reasoning components, and a dedicated SKAB validation path.

Overview

ANSR-DT integrates three main elements:

• CNN-LSTM based temporal pattern learning for anomaly detection.
• Prolog-based symbolic reasoning for explicit rules, explanations, and rule updates.
• PPO-based adaptation for downstream control and policy refinement.

The project supports two complementary evaluation settings:

• A synthetic digital-twin pipeline used for controlled end-to-end ANSR-DT experiments.
• A dedicated SKAB-native path used for real-world benchmark validation on industrial sensor streams.

Reported Results

ANSR-DT is evaluated in both controlled synthetic digital-twin experiments and a dedicated SKAB-based real-world benchmark setting. The table below summarizes the main reported outcomes.

Setting	What ANSR-DT demonstrates	Representative outcome
Synthetic digital-twin benchmark	Strong neuro-symbolic anomaly detection with interpretable reasoning traces	F1 0.966, ROC-AUC 0.955
Symbolic scalability	Rule-based inference remains practical as the rule base grows	100 rules with sub-6 ms latency
Real-world SKAB validation	The ANSR-DT design transfers to realistic industrial sensor streams through a dedicated SKAB-native adaptation	Symbolic SKAB variant: F1 0.755, ROC-AUC 0.859
Overall contribution	ANSR-DT combines prediction, symbolic traceability, and adaptive decision support in one framework	Competitive performance with explicit rules and benchmark-based external validation

Architecture

ANSR-DT is organized around a physical sensing layer, a neuro-symbolic processing layer, and an adaptation layer. The core pipeline links multivariate sensor windows to neural anomaly scores, symbolic facts and rules, fused decisions, and optional control actions.

Installation

Prerequisites

• Python 3.8+
• Anaconda or another virtual-environment manager
• SWI-Prolog
• ProbLog

Setup

git clone https://github.com/sbhakim/ansr-dt.git
cd ansr-dt
conda create -n ansr_dt_env python=3.9
conda activate ansr_dt_env
pip install -r requirements.txt
pip install problog

Quick Start

Core ANSR-DT pipeline

python main.py

Dedicated SKAB pipeline

python -m src.skab.run --config configs/config_skab_separate.yaml

The curated local SKAB subset used by this repository is stored under data/SKAB. That directory includes only the CSV files used by the ANSR-DT SKAB pipeline plus a short local dataset note with external source links.

Repository Layout

.
├── configs/
│   ├── config.yaml
│   ├── config_skab.yaml
│   └── config_skab_separate.yaml
├── data/
│   ├── SKAB/
│   └── synthetic_sensor_data_with_anomalies.npz
├── main.py
├── src/
│   ├── ansrdt/
│   ├── data/
│   ├── evaluation/
│   ├── inference/
│   ├── models/
│   ├── pipeline/
│   ├── reasoning/
│   ├── rl/
│   ├── skab/
│   ├── training/
│   └── utils/
└── requirements.txt

Publications

Primary manuscript

@article{hakim2025ansr,
  title={ANSR-DT: An Adaptive Neuro-Symbolic Learning and Reasoning Framework for Digital Twins},
  author={Hakim, Safayat Bin and Adil, Muhammad and Velasquez, Alvaro and Song, Houbing Herbert},
  journal={arXiv preprint arXiv:2501.08561},
  year={2025}
}

Supplementary associated paper

S. B. Hakim, M. Adil, A. Velasquez and H. H. Song, "An Explainable Neuro-Symbolic Rule Extraction Framework for Digital Twins," 2025 IEEE Smart World Congress (SWC), Calgary, AB, Canada, 2025, pp. 1042-1047.
DOI: 10.1109/SWC65939.2025.00168

License

This project is licensed under the MIT License. See LICENSE.

Contact

Please open an issue for project questions or contact Safayat Bin Hakim at safayat DOT b DOT hakim AT gmail DOT com.