DAYPSCI: Event-based dataset for anomaly detection through fault injection in PLC-controlled industrial cyber-physical systems
TL;DR
DAYPSCI is presented, an event-based dataset generated using an industrial cyber-physical system (CPS) testbed based on a PLC-controlled part marking station with Siemens S7-1200 and S7-1500 devices that supports the development and evaluation of machine learning methods for anomaly detection and fault classification in industrial CPS.
Abstract
This article presents DAYPSCI, an event-based dataset generated using an industrial cyber-physical system (CPS) testbed based on a PLC-controlled part marking station with Siemens S7-1200 and S7-1500 devices. The system integrates real industrial hardware with digital twin technologies, enabling controlled and repeatable experiments. Data acquisition follows an event-based logging approach, where only changes in system variables are recorded rather than using fixed sampling rates, and each event is associated with its corresponding inter-event time (Δt), enabling precise temporal characterization of system dynamics. The dataset also includes scan-level identifiers (scan_id) and event ordering (event_order), preserving the logical execution order within PLC scan cycles. It contains time-stamped records of digital sensors, solenoid valve control signals, and process states under both normal operation and controlled fault injection scenarios affecting sensors, actuators, or both. Ground truth is generated through a hybrid approach combining externally defined labels from the experimental configuration and labels derived at runtime from control system signals (e.g., GEMMA states), ensuring a clear separation between CPS execution and the labeling layer while enabling traceability between injected faults (cause) and observable system behavior (effect), and allowing differentiation between fault activation and observable anomaly manifestation, which may be temporally decoupled. The dataset is organized into independent experimental batches, each including processed data (CSV), network traffic captures (PCAPNG) from a Profinet-based industrial communication network, and detailed documentation, facilitating sequence-based analysis and reproducibility. The dataset supports the development and evaluation of machine learning methods for anomaly detection and fault classification in industrial CPS.