DRL-Based Decision Support

Welcome to the Control Room Simulation Dataset repository! This dataset contains valuable information derived from a simulated formaldehyde production plant, involving participant interaction within a controlled experimental setting resembling a control room. The dataset is utilized for statistical analysis to compare outcomes among different groups and has potential applications for decision-makers involved in control room design and optimization, process safety engineers, system engineers, human factors engineers in process industries, and researchers in related domains. This repository contains the code for DRL-based decision support tool usage in control room scenarios:

https://arxiv.org/abs/2402.13219

Overview

The dataset encompasses measurements obtained from diverse data sources during the period spanning May to August 2023. It incorporates both objective and subjective metrics commonly employed in assessing cognitive states related to workload, situational awareness, stress, and fatigue. Various data collection tools such as health monitoring watch, eye tracking, process and HMI logs, operational metrics (response time, reaction time, performance, etc), NASA Task Load Index (NASA-TLX), Situation Awareness Rating Technique (SART), a think-aloud Situation Presence Assessment Method (SPAM), AI support questions, and AI vs human error were utilized.

Data Collection

Participants tested three scenarios lasting 15–18 minutes, with breaks and survey completion periods in between, utilizing different combinations of decision support tools. The decision support tools varied across groups, encompassing factors of digitized screen-based procedures and the inclusion of an AI recommendation system.

Dataset Structure

The collected raw data was processed particularly for the analyses in the associated research paper. The data from individual participants was concatenated and merged into a single xls file for further evaluation. The data used for comparison between the GroupN and GroupAI is presented in the merged normalized data folder. The xls file contains the data points for each participant per row, and each column represents the data and sub-data collected from various sources. As the focus of this analysis is not between the scenarios, the data is normalized across scenarios to avoid the effect of the scenarios in the analysis and is averaged for every participant to acquire a single vector of data for each participant.

Merged Normalized Data

• File Name: merged_normalized_data.xls
• Description: This file contains the merged and normalized data for comparison between GroupN and GroupAI. Each row represents a participant, and each column contains the data and sub-data collected from various sources. The data is normalized across scenarios to avoid the effect of the scenarios in the analysis and is averaged for every participant to acquire a single vector of data for each participant.

Data Columns

1. Biometric Measures

   • Electrodermal Activity (EDA) or Galvanic Skin Response (GSR): Measures the skin’s electrical conductance, influenced by sweat gland activity.
   • Pulse Rate or Heart Rate: Defines the number of heartbeats per unit time (bpm).
   • Temperature: Represents the body temperature, indicating the degree of coldness or hotness of the body.

2. Operational Measures

   • Response Time: The time taken to respond to a specific event or stimulus.
   • Reaction Time: The time taken to initiate a response once a stimulus is presented.
   • Performance Metrics: Various metrics related to the performance of the participants during the simulation tasks.

3. AI Support Metrics

   • AI Acknowledgement: The number of times the AI acknowledgment button was pressed during the duration of each scenario.
   • AI vs Human Response: Deviations and preliminary analysis of decisions taken by the human participant and suggestions by AI/DRL agent.

4. Process, Alarms, and HMI Logs

   • Number of Alarms Annunciated: The count of alarms announced during the simulation.
   • Number of Alarms Silenced: The count of alarms silenced by the participants.
   • Number of Alarms Acknowledged: The count of alarms acknowledged by the participants.
   • Number of Procedures Opened: The count of procedures opened during the duration of each scenario.
   • Number of Mimics Opened: The count of mimics opened during the duration of each scenario.

Value of the Data

The dataset provides an opportunity to study the integration of human-in-the-loop configurations with AI systems in safety-critical industries. By examining the data, researchers can identify the factors necessary for successful collaboration between humans and AI. This knowledge can lead to the development of optimized interaction mechanisms, ensuring that the strengths of both humans and AI are leveraged effectively to enhance decision-making in critical scenarios.

Accessing the Dataset

The dataset files are organized in the following structure:

• merged_normalized_data.xls: Contains the merged and normalized data for comparison between GroupN and GroupAI.
• hmm_modeling_concatenated_data.csv: Represents the time-series data of the process, alarms, and HMI logs for every participant into a single file as required by the HMM python library.
• failed_participants_labels.csv: Provides the labels for participants who failed during the task based on various factors such as the consequence of plant shutdown or reactor overheating and overall performance.

Citation

If you use this dataset in your research or publication, please cite the associated research paper.

We hope this dataset proves to be valuable for your research and analysis. If you have any questions or need further assistance, feel free to reach out to the repository maintainers. Thank you for your interest in the Control Room Simulation Dataset!