Predicting the next event during process runtime is an objective of interest in predictive process monitoring (PPM).Decay replay miningis one of few deep learning-based next event prediction approaches that are built upon process model notations. However, this algorithm does not fully intertwine its neural network with the available process knowledge contained in the process model. This work, which is an extended version of an earlier conference publication, investigates the reachability graphs of underlying Petri net process models for masking the neural network ofdecay replay miningto ultimately increase the quality of next event predictions. A more comprehensive set of experiments is performed to provide robust statistical evidence of the usefulness of the approach and relativizes earlier made claims and hypotheses. In addition, thedecay replay miningapproach is applied with the suggested reachability graph-based masking extension to a healthcare use case of sepsis patients facilitating decision-making for healthcare practitioners. The obtained results further underscore the validity of the masking of neural networks using knowledge contained in the reachability graph of a Petri net process model.

Diabetes intensive care unit (ICU) patients are at increased risk of complications leading to in-hospital mortality. Assessing the likelihood of death is a challenging and time-consuming task due to a large number of influencing factors. Healthcare providers are interested in the detection of ICU patients at higher risk, such that risk factors can possibly be mitigated. While such severity scoring methods exist, they are commonly based on a snapshot of the health conditions of a patient during the ICU stay and do not specifically consider a patient's prior medical history. In this paper, a process mining/deep learning architecture is proposed to improve established severity scoring methods by incorporating the medical history of diabetes patients. First, health records of past hospital encounters are converted to event logs suitable for process mining. The event logs are then used to discover a process model that describes the past hospital encounters of patients. An adaptation of Decay Replay Mining is proposed to combine medical and demographic information with established severity scores to predict the in-hospital mortality of diabetes ICU patients. Significant performance improvements are demonstrated compared to established risk severity scoring methods and machine learning approaches using the Medical Information Mart for Intensive Care III dataset.

Time series classification models have been garnering significant importance in the research community. However, not much research has been done on generating adversarial samples for these models. These adversarial samples can become a security concern. In this paper, we propose utilizing an adversarial transformation network (ATN) on a distilled model to attack various time series classification models. The proposed attack on the classification model utilizes a distilled model as a surrogate that mimics the behavior of the attacked classical time series classification models. Our proposed methodology is applied onto 1-nearest neighbor dynamic time warping (1-NN DTW) and a fully convolutional network (FCN), all of which are trained on 42 University of California Riverside (UCR) datasets. In this paper, we show both models were susceptible to attacks on all 42 datasets. When compared to Fast Gradient Sign Method, the proposed attack generates a larger faction of successful adversarial black-box attacks. A simple defense mechanism is successfully devised to reduce the fraction of successful adversarial samples. Finally, we recommend future researchers that develop time series classification models to incorporating adversarial data samples into their training data sets to improve resilience on adversarial samples.

This work-in-progress research paper explores the way in which low-socioeconomic status (SES), first-year undergraduate engineering students develop their engineering identity. Identification with the field of engineering, or engineering identity development, is an ongoing process for students. While scholars have used retrospective studies to understand the developmental aspect of this process, a longitudinal study that follows students' engineering identity development could provide an advantageous viewpoint. In this study, we investigate the engineering identity profiles of incoming low-SES, high-achieving engineering students. We interviewed 13 students using a protocol focused on understanding the students' engineering identity profiles before entering engineering school. An integrated model of engineering identity development was used to frame the research and guide the analysis. Our preliminary results show existing pre-college identity-related patterns across students as well as initial ways of identifying with their major and engineering as a field. This work has contributions to research in the areas of engineering identity development as well as broadening understanding of engineering students who are both low-income and high-achieving. Our work has practical implications for academic and professional support programs for low-income engineering students and programs that aim to support engineering identity development.

Process Mining is a famous technique which is frequently applied to Software Development Processes, while being neglected in Human-Computer Interaction (HCI) recommendation applications. Organizations usually train employees to interact with required IT systems. Often, employees, or users in general, develop their own strategies for solving repetitive tasks and processes. However, organizations find it hard to detect whether employees interact efficiently with IT systems or not. Hence, we have developed a method which detects inefficient behavior assuming that at least one optimal HCI strategy is known. This method provides recommendations to gradually adapt users' behavior towards the optimal way of interaction considering satisfaction of users. Based on users' behavior logs tracked by a Java application suitable for multi-application and multi-instance environments, we demonstrate the applicability for a specific task in a common Windows environment utilizing realistic simulated behaviors of users.

Vocal disorders have affected several patients all over the world. Due to the inherent difficulty of diagnosing vocal disorders without sophisticated equipment and trained personnel, a number of patients remain undiagnosed. To alleviate the monetary cost of diagnosis, there has been a recent growth in the use of data analysis to accurately detect and diagnose individuals for a fraction of the cost. We propose a cheap, efficient and accurate model to diagnose whether a patient suffers from one of three vocal disorders on the FEMH 2018 challenge.

Throughout recent years, dynamic time warping (DTW) has remained as a robust similarity measure in time series classification (TSC). 1-nearest neighbor (1-NN) algorithm with DTW is the most widely used classification method on time series serving as a benchmark. With the increasing demand for TSC on low-resource devices and the widespread of wearable devices, the need for a efficient and accurate time series classifier has never been higher. Although 1-NN DTW attains accurate results, it highly falls back on efficiency due to its quadratic complexity in the length of time series. In this paper, we propose a new approximation method for reducing the length of the time series as the input of DTW. We call it control chart approximation (CCA), after a similar concept used in statistical quality control processing. CCA representation approximates raw time series by transforming them into a set of segments with aggregated values and durations forming a reduced 3-D vector. We also propose an adaptation of DTW in 3-D space as a distance measure for 1-NN classifier, and denote the method as 1-NN 3-D DTW. Our experiments on 85 datasets from UCR archive-including 28 long-length (>500 points) time series datasets-show up to two orders of magnitude performance gain in running time compared to the state-of-the-art 1-NN DTW implementation. Moreover, it shows similar or better accuracy on the long time series in the experiment.

Predicting students' academic performance (SAP) provides invaluable information for educational institutes' authorities. This information offers numerous opportunities for instructors and decision makers to improve their quality of services and consequently help the students to succeed in their education. In this paper, we introduce a prediction model to forecast the SAP of the Engineering students. The model is based on the Bayesian networks framework. The model is constructed using a database of the undergraduate engineering students at University of Illinois at Chicago (UIC). The specific objective of this model is to predict the students' grades in three major courses which most of the students take in their second semester. The grades in these courses have major impact on students' retention rates as many students receive low grades in them. Therefore, predicting students' grades in these courses can be used to identify the students who might receive low grades and hence need extra help from the educational authorities. The proposed model has been tested against the conventional models which have been proposed in the literature and it is proven to outperform them in grade prediction.

We define a method for the automatic generation of supervisory controllers that force a plant to perform a given operation by a given deadline. The operation must be executed by a prespecified delay λ with respect to the previous execution of the operation. Although our supervisor generation occurs offline with respect to plant execution, the resulting controllers automatically take into account variable task durations in an effort to increase the flexibility of operation schedules in the controlled plant. We model both the controlled plant and control supervisors as time Petri nets. In this setting, our control supervisors must force a target transitiontdto fire within λ time units since the previous firing oftd. Our supervisor generation is based on the concept of a transition latency. The latency of a Petri net transitiontis the time interval during whichtmust be disabled in order for target transitiontdto fire by its deadline. If a transitiontthat may delay the firing oftdhas latencyl(t), thentmust be disabled at leastl(t) time units before the expiration of the deadline ontd. In this paper, we discuss in detail two algorithms for generating transition latencies, and we show an application to a maintenance system.

This paper defines a reconfiguration method for the class of discrete-event systems (DES) that is subject to linear constraints as their control specifications. Some existing methods for enforcing these constraints make use of Petri-net P-invariants for controller synthesis. These methods are quite appealing because their computational complexity is much more tractable than most other methods for controller synthesis. However, a common limitation of all existing P-invariant-based control architectures for DES plants is the assumption that the linear constraints defining the control specification of the plant do not change over time. Here, we relax this assumption and allow the control specifications to change during controller runtime. Under certain assumptions on DES behavior, we automatically reconfigure the DES controller after the control specification is changed. In addition, if the current state of the controlled DES has become infeasible under the new control specification, we automatically generate a so-called plant reconfiguration procedure whose execution leads the system back to a feasible state. This reconfiguration procedure is optimal in that it seeks to minimize the cost of reconfiguration actions through an Integer Programming (IP) model. The objective function of the IP model can be used to generate reconfiguration solutions that meet some desired properties. Depending on the cost of each reconfiguration action, a minimum cost reconfiguration solution may use only actions contained in the current plant configuration (an internal response), or ask for a change in the plant configuration, for instance, by adding new resources (an external response), or a combination of both strategies. Finally, we illustrate our method by applying it to a hospital control system example. Note to Practitioners-This paper proposes a dynamic reconfiguration framework that can revise the operations of systems whose control requirements change over time. The proposed framework can be applied to systems that satisfy the following two assumptions. First, the behavior of the system under study is described in terms of a set of discrete states and events. Events will cause the system to transition between states. Second, the control requirements must be expressed by linear equalities and inequalities on the system states. Under these circumstances, the proposed framework can identify an optimal transition to a new control policy that satisfies the new control requirements. Moreover, the system under consideration will continue operating while this transition is taking place. One application of this method is in modifying hospital control strategies when a hospital experiences unexpected events. In this case, the hospital operations-such as patient handling, resource assignment, and procedure scheduling-can be represented by discrete state models (e.g., Petri nets). Constraints on these operations can be modeled by linear inequalities on hospital and patient state. Upon a change in the constraints, the proposed reconfiguration method revises the hospital control strategies. For example, a shift in the hospital service demands (e.g., an increase in the flow of patients to the hospital due to a mass casualty situation) can be translated to changes in the constraints. In this case, the hospital operations must be revised to accommodate the new constraints without disrupting the operation of the hospital. The reconfiguration method of this paper provides a framework for modeling the reconfiguration steps and for calculating the least cost reconfiguration solution.

We automatically synthesize supervisory controllers that force a system to perform a certain operation by a given deadline. The operation must be executed by a pre-specified delay λ with respect to the previous execution of the operation. We model both the controlled system and our control supervisors as time Petri nets. Given a target transition and a deadline, our supervisors disable net behaviors in which the firing of the target transition may miss the deadline. Our method is subject to a merge exclusion assumption on the structure of paths contained in a so-called net unfolding. If a control problem does not satisfy this assumption, we abandon supervisor generation. Preliminary empirical results show that our method is both relatively general and more tractable from a computational standpoint than most other real-time analysis methods.

Wireless indoor positioning systems have become very popular in recent years. These systems have been successfully used in many applications such as asset tracking and inventory management. This paper provides an overview of the existing wireless indoor positioning solutions and attempts to classify different techniques and systems. Three typical location estimation schemes of triangulation, scene analysis, and proximity are analyzed. We also discuss location fingerprinting in detail since it is used in most current system or solutions. We then examine a set of properties by which location systems are evaluated, and apply this evaluation method to survey a number of existing systems. Comprehensive performance comparisons including accuracy, precision, complexity, scalability, robustness, and cost are presented.

We define supervisory controllers for enforcing deadlines on transition firings in time Petri nets. Given a target net transition t/sub d/, and a deadline /spl lambda/, we generate a controller that under broad assumptions forces t/sub d/ to fire every /spl lambda/ time units. Our supervisory controller is based on the notion of transition latency. The latency of a net transition is an upper bound on the time between the firing of that transition and the firing of t/sub d/. A transition is not allowed to fire when its latency is greater than the amount time left until t/sub d/ must fire. Our real-time supervisory controllers consist of two subnets, which are added to the controlled net in order to enforce deadline /spl lambda/ on the firing of t/sub d/. First, the clock subnet dynamically tracks the amount of time left until the expiration of /spl lambda/. As the deadline approaches, this subnet also indicates transitions that must be disabled because their latency has become greater than the time until the expiration of the deadline. When this happens, a supervisor subnet actually disables these transitions. These transitions are enabled again only after t/sub d/ has fired.

We outline a supervisory control method for enforcing deadlines in real-time systems modeled as time Petri nets. Given a time Petri net NscrT, a net transition td, and a deadline d, we generate supervisory controllers that force tdto fire at most d time units since the previous firing of td. Our method uses the unfolding of the ordinary (untimed) Petri net underlying NscrTin order to define control supervisors enforcing the deadline. The unfolding of a Petri net is an acyclic Petri net whose markings correspond to the markings of the original Petri net. An advantage of net unfoldings is that they show the causal relation among transition firings in the original Petri net. Using a net unfolding, we define a so-called latency for each transition in NscrT, which we assume to be live. The latency of a transition is an upper bound on the delay between the firing of that transition and the firing of td. Our supervisory controllers dynamically disable the firing of transitions whose latency is greater than the time left until the expiration of the deadline on td. Here we discuss the most crucial aspect of this method, namely the definition of transition latencies from net unfoldings

In this paper, we develop a framework for reconfiguration of a discrete event system (DES) controller, which has a dynamic event observation set. We will show the designed reconfiguration yields a more tolerable controller than the one designed in [5]. Starting with a maximally permissive controller that has a full observation of its DES, we design a mega-controller that monitors the observation set of the DES controller and its state continuously. Upon a change in the observation set, the mega-controller reconfigures the controller by a aggregation or disaggregation of the controller states. The mega-controller is also responsible for feedback function adjustments if the available observation set causes a conflict in control. We illustrate the reconfiguration procedures by an example.

In this paper, we introduce the concept of switching from one supervisory control to another one when changes in the state of an embedded sensory network occur. The sensory devices are assumed to be single functional, that is, they are programmed to observe and report only one event. The switching model is built based on the two criteria of language observability and state synchronization. Switching model is the basis for a megacontroller, which monitors the state of the sensory network embedded in the plant as well as the state of the supervisor in charge. Every time the state of this network changes, the megacontroller reconfigures the control system by finding an appropriate supervisor.

Despite the efforts in scheduling and control of flexible manufacturing systems (FMSs) with resource constraints, the current pool of scheduling techniques faces two major drawbacks: modeling and complexity. Modeling is the task of converting the FMS data to a set of information, ready to be processed by a scheduling algorithm. Complexity has a direct relation with the amount of effort required to execute a scheduling algorithm successfully on the information set generated in the modeling phase. In this paper, we use finite automata (FA) theory to develop a modeling formalism and its accompanying scheduling algorithm for control and scheduling of FMS with resource constraints. While the FA-based modeling is completely automatic and does not need any human-designer interference, its related algorithm is both effective and efficient. We use IDEF3 standard to capture the FMS activities and resource data. We propose a three-step procedure. In the first step, the IDEF3 data set is converted to a finite automaton, preserving the activity precedence relationships. In the second step, the resulted finite automaton is decomposed to smaller (in size) scheduling problems that can be independently optimized. In the third step, a heuristic scheduling algorithm is used to handle each problem separately. We applied the developed procedure to 100 problems. The results are satisfactory and promising.

We used Ramadge-Wonham (RW) theory (1987) of supervisory control to control a system of mobile robots. We discuss our experience in modeling and implementation of the developed control system. We specifically address the control program structure that manages the interaction of the RW controller with its plant. We also present our approach in dealing with practical issues such as forcing events and simultaneous events. The advantages and disadvantages of the RW controller are discussed.

We present a supervisory control architecture with control redundancy. We assume an adaptive architecture for the problem of switching from one control module to another when sensors used for observing events in the plant fail or become unavailable. It is assumed that there is an actual controller and a number of virtual controllers. The actual controller is in real feedback loop with the plant whereas the virtual ones are in virtual feedback such that the coupled behavior of any of these control modules with plant conforms to what is called normality criterion in the context of partial observation. Every time that a failure occurs and the environment changes, the actual controller is replaced by one of these virtual ones. The selection is based on some optimality criterion and user defined constraints.

The Ramadge-Wonham (RW) framework for designing a supervisory controller for a given plant and user specification, has been widely addressed by many researchers. The framework accepts a generator (or plant model), G, and a specification (or imposed constraint on the plant), H, and generates the supervisory controller, S. It assumes that plant is known and deterministic. However, in many practical cases the designer is not completely aware of the real generator G/sub R/. In this paper the above assumption is removed. It is assumed that G is not fully known. Using the available G and the user specification H, the RW procedure is applied to a supervisor S. Then the family of (real) generators, {G/sub R/}, which can be supervised under S while satisfying the specification H is introduced and different cases are discussed. We establish the mathematical framework for one of the cases where some events originally controllable become uncontrollable or vice versa. It is argued that the complement of {G/sub R/} which violates the specification H can be partially synthesized by available recovery and diagnosis techniques in RW literature if the violation is already known. Otherwise the violated specification must be handled in run time.