Babji Srinivasan

The objective of the design and operation of any water distribution network (WDN) includes meeting the desired demand at sufficient pressure at all nodes. However, this requires situational awareness; in other words, the knowledge of system state variables such as pressure and flow throughout the network. In this work, a hybrid approach is developed for sensor placement (SP) and state estimation (SE) that exploits the underlying correlation structure in the data, along with the principles governing the flow through circular pipes. The problem of SP in WDN is addressed since measuring the state variables throughout the network is not practical. The problem of SE that maps to a matrix completion problem under certain physical and logical constraints is solved later. The completed matrix represents the state of WDN at any given time. Benchmark networks used in literature were used to evaluate the proposed approach. The mean absolute percentage error (MAPE) of less than 5% was obtained while estimating the head available at nodes. The knowledge of the states in the entire network could help operate the network adaptively.

Events such as pipe burst, pipe clogging, excess demand often disrupt the water distribution network operation. As a result, water utilities have to bear an extra financial burden. The knowledge of such events could help the water supply board formulate and operate the network with minimal loss. This work aims to identify the nature of such uncertain events with limited operational information using simple measures designed to monitor the network locally at nodes. Instead of acquiring information from all the nodes, a few nodes are identified for sensor placement by solving a Mixed Integer Linear Programming (MILP) problem. The objectives for solving MILP are to minimise total sensor cost and maximise the sensitivity of measurements to any given fault. Sensitivity is calculated from operational data generated by performing Monte Carlo simulations. These simulations generate data for each fault of different magnitude at different locations. Gradient boosting trees are trained using the limited operational information from the network. Accuracy of identification of faults of 78% on the passive network and over 81% on active networks was obtained using limited features with extreme gradient boosting (XGBoost) model.

Fouling degrades the overall efficiency of the heat exchanger networks (HENs), which results in a significant economic loss. The mitigation of fouling in an operational HEN is carried out by optimizing the cleaning schedules of the heat exchangers. Although such approach can save costs, it is subjected to the exact implementation of the optimal cleaning schedule. Usually, the small and medium-scale process industries face difficulties in implementing such solutions due to limited resources, which forces them to rely on suboptimal cleaning schedules, such as postponing or avoiding few cleaning tasks. This work addresses this gap by optimizing the cleaning schedule considering the maintenance resource limitation. Our approach considers a mixed-integer linear programming (MILP) based optimization considering groupings of heat exchangers based on their spatial locations for ease of maintenance The proposed formulation is applied on a HEN with linear and asymptotic fouling, with and without cleaning cost. The results show that the approach can prevent a considerable economic loss, which would incur due to suboptimal cleaning schedules due to resource limitations.

Traditional matrix and tensor completion approaches utilize latent structures in data to impute missing entries. Recent works on distribution system state estimators employing such data imputation techniques have identified the need to incorporate fundamental system equations as constraints to improve state estimation accuracy. As a result, these techniques provide superior state estimation performance compared to their model-free counterparts and conventional state estimators. In practice, the data required for these estimators are provided by sensors/meters deployed in the network. However, prior efforts do not explore the placement of sensors that optimize the performance of the estimators. Moreover, constraints on entries of these matrices and tensors result in specific combinations of known measurements to provide better imputation results than others. Therefore, this work proposes two-meter placement approaches that leverage network parameters and linearized power flow equations to identify sensor locations. These approaches achieve this by iteratively placing sensors with the highest contribution towards minimizing the voltage residual in the selected reference cases. The first approach identifies buses that provide the highest reduction in voltage residuals. In contrast, the second approach identifies locations of a heterogeneous set of sensors that provide the highest reduction in voltage residual. The proposed approaches can also extend existing sensor deployments such as distribution phasor measurement units (D-PMU) and supervisory control and data acquisition (SCADA) sensing and measurement devices (e.g., Bellwether meters) to improve state estimation performance. The approaches have been evaluated on the IEEE 33-node distribution system and an unbalanced 3-phase 559-node distribution system.

Operators' mental models play a central role in safety-critical domains like the chemical process industries. Accurate mental models, i.e., a correct understanding of the process and its causal linkages, are prerequisites for safe operation. Mental models are often defined and explained in abstract terms that make their interpretation subjective and prone to bias. In this work, we propose a Hidden Markov Model (HMM) based formalism to characterize control room operators' mental models while handling abnormal situations. We show that a suitable HMM representing the operator's mental model – including the states, state transition probabilities, and emission probability distributions – can be identified experimentally using data of the operator's control actions, eye gaze, and process variable values. This HMM can be used for the quantitative assessment of operators' mental models as illustrated using various case studies. We discuss the potential applications of the model in identifying various cognitive errors and human reliability assessment. In Part 2 of this paper, we use the proposed approach to assess operators' learning during training.

Process plant operators rely on their knowledge of process cause-and-effect relationships during abnormal situation management. Novice operators develop such process knowledge during training. Hence, holistic assessment of operators’ training is essential to ensure process safety. Currently, during training, operators’ process understanding is evaluated using criteria such as successful completion, task based measures, and operator actions that ignore their cognitive behavior. In this work, we propose an eye-tracking-based approach that uses the operator's attention allocation during different pre-specified training scenarios along with process data, alarm information, and operator actions. Our approach is based on the precept that an operator would focus their attention on those variables on the human-machine interface that they believe have a direct causal relationship to the situation at hand. Also, expert operators seek time-based information for proactive monitoring. Accordingly, to quantify the progress of a novice operator's learning, we develop two metrics — association metric and salience metric — using correspondence analysis of operators' eye gaze. To evaluate the applicability of the metrics, we conducted experiments with ten participants who performed 486 tasks. Statistical studies reveal that the proposed metrics can quantify operators’ learning and thus can be used to objectively evaluate training effectiveness.

To ensure safe and efficient operation, operators in process industries have to make timely decisions based on time-varying information. A holistic assessment of operators’ performance is, therefore, challenging. Current approaches to operator performance assessment are subjective and ignore operators’ cognitive behavior. In addition, these cannot be used to predict operators’ expected responses during novel situations that may arise during plant operations. The present study seeks to develop a human digital twin (HDT) that can simulate a control room operator’s behavior, even during various abnormal situations. The HDT has been developed using the ACT-R (Adaptive Control of Thought-Rational) cognitive architecture. It mimics a human operator as they monitor the process and intervene during abnormal situations. We conducted 426 trials to test the HDT’s ability to handle disturbance rejection tasks. In these simulations, we varied the reward and penalty parameters to provide feedback to the HDT. We validated the HDT using the eye gaze behavior of 10 human subjects who completed 110 similar disturbance rejection tasks as that of the HDT. The results indicate that the HDT exhibits similar gaze behaviors as the human subjects, even when dealing with abnormal situations. These indicate that the HDT’s cognitive capabilities are comparable to those of human operators. As possible applications, the proposed HDT can be used to generate a large database of human behavior during abnormalities which can then be used to spot and rectify flaws in novice operator’s mental models. Additionally, the HDT can also enhance operators’ decision-making during real-time operation.

Process industries are highly hazardous, and these hazards often lead to accidents. Over 70% of these accidents are attributed to human errors. With the advancements in technology and changing role of operators to the one involving an emphasis on cognitive aspects, most of these errors occur due to limitations in cognitive performance. One of the major constructs to understand cognitive performance is the cognitive workload. An increase in cognitive workload often leads to degradation in performance. Eye tracking has been used in several domains to assess cognitive workload. In this work, we propose a methodology to assess cognitive workload of control room operators during tasks that involve tackling process abnormalities. The methodology employs the fusion of metrics obtained from pupil and gaze data. Our results reveal that fusion of metrics provides better accuracies of classifying cognitive workload at three levels—low, medium and high workload.

Internet of Things (IoT) is a key enabler of Industry 4.0 with networked devices providing sensor data to help manage, automate, streamline and optimize assets, operations and processes. In such industrial IoT settings, reliability and process experts spend a considerable amount of time in creating accurate ground-truth data to assist with the inferencing capabilities of Artificial Intelligence (AI) engines. This process can be time-consuming and sometimes inaccurate, depending on the complexity of data. Accurate expert annotated data is the foundation for many AI applications because data needs to be classified on several bases, for instance into 'normal', 'abnormal' or 'pre-abnormal' states. Such problem formulations can be appropriately addressed using Active Learning (AL) techniques. We propose an AL framework capable of handling two practical challenges: oracle uncertainty and quantification of model performance in the absence of ground truth. Consequently, the proposed approach addresses uncertainties within AL techniques by fusing information pertaining to expertise levels of the human annotators and their confidence levels corresponding to the annotation provided.

Human performance is critical to ensure safety and health in hazardous chemical settings, wherein the majority of the accidents have been attributed to human error. As a result, both industry and academia have been concerned about the efficacy of knowledge and skill transfer. Virtual reality (VR) technology is gaining attention as a way to improve human performance. In this paper, we review the literature on the applications of VR to chemical safety in laboratories and industries. Our review reveals that VR offers much potential for training lab users and plant operators as well as for bridging the theoretical knowledge-practical skills gap. However, there is a need to develop systematic approaches to measure the effectiveness in achieving the desired training outcomes. Therefore, in this paper, we discuss the best practices for VR-based training. We also stress the need to incorporate physiological sensors into the VR environment.