Evaluating Technicians' Workload and Performance in Diagnosis for Corrective Maintenance.

The advancement in digital technology is transforming the world. It enables smart product-service systems that improve productivity by changing tasks, processes, and the ways we work. There are great opportunities in maintenance because many tasks require physical and cognitive work, but are still carried out manually. However, the interaction between a human and a smart system is inevitable, since not all tasks in maintenance can be fully automated. Therefore, we conducted a controlled laboratory experiment to investigate the impact on technicians' workload and performance due to the introduction of smart technology. Especially, we focused on the effects of different diagnosis support systems on technicians during maintenance activity. We experimented with a model that replicates the key components of a computer numerical control (CNC) machine with a proximity sensor, a component that requires frequent maintenance. Forty-five participants were evenly assigned to three groups: a group that used a Fault-Tree diagnosis support system (FTd-system), a group that used an artificial intelligence diagnosis support system (AId-system), and a group that used neither of the diagnosis support systems. The results show that the group that used the FTd-system completed the task 15% faster than the group that used the AId-system. There was no significant difference in the workload between groups. Further analysis using the NGOMSL model implied that the difference in time to complete was probably due to the difference in system interfaces. In summary, the experimental results and further analysis imply that adopting the new diagnosis support system may improve maintenance productivity by reducing the number of diagnosis attempts without burdening technicians with new workloads. Estimates indicate that the maintenance time and the cognitive load can be reduced by 8.4 s and 15% if only two options are shown in the user interface.

Leveraging Active Learning for Failure Mode Acquisition.

Identifying failure modes is an important task to improve the design and reliability of a product and can also serve as a key input in sensor selection for predictive maintenance. Failure mode acquisition typically relies on experts or simulations which require significant computing resources. With the recent advances in Natural Language Processing (NLP), efforts have been made to automate this process. However, it is not only time consuming, but extremely challenging to obtain maintenance records that list failure modes. Unsupervised learning methods such as topic modeling, clustering, and community detection are promising approaches for automatic processing of maintenance records to identify failure modes. However, the nascent state of NLP tools combined with incompleteness and inaccuracies of typical maintenance records pose significant technical challenges. As a step towards addressing these challenges, this paper proposes a framework in which online active learning is used to identify failure modes from maintenance records. Active learning provides a semi-supervised machine learning approach, allowing for a human in the training stage of the model. The hypothesis of this paper is that the use of a human to annotate part of the data and train a machine learning model to annotate the rest is more efficient than training unsupervised learning models. Results demonstrate that the model is trained with annotating less than ten percent of the total available data. The framework is able to achieve ninety percent (90%) accuracy in the identification of failure modes in test cases with an F-1 score of 0.89. This paper also demonstrates the effectiveness of the proposed framework with both qualitative and quantitative measures.

Digital Twin for Training Bayesian Networks for Fault Diagnostics of Manufacturing Systems.

Smart manufacturing systems are being advocated to leverage technological advances that enable them to be more resilient to faults through rapid diagnosis for performance assurance. In this paper, we propose a co-simulation approach for engineering digital twins (DTs) that are used to train Bayesian Networks (BNs) for fault diagnostics at equipment and factory levels. Specifically, the co-simulation model is engineered by using cyber-physical system (CPS) consisting of networked sensors, high-fidelity simulation model of each equipment, and a detailed discrete-event simulation (DES) model of the factory. The proposed DT approach enables injection of faults in the virtual system, thereby alleviating the need for expensive factory-floor experimentation. It should be emphasized that this approach of injecting faults eliminates the need for obtaining balanced data that include faulty and normal factory operations. We propose a Structural Intervention Algorithm (SIA) in this paper to first detect all possible directed edges and then distinguish between a parent and an ancestor node of the BN. We engineered a DT research test-bed in our laboratory consisting of four industrial robots configured into an assembly cell where each robot has an industrial Internet-of-Things sensor that can monitor vibrations in two-axes. A detailed equipment-level simulator of these robots was integrated with a detailed DES model of the robotic assembly cell. The resulting DT was used to carry out interventions to learn a BN model structure for fault diagnostics. Laboratory experiments validated the efficacy of the proposed approach by accurately learning the BN structure, and in the experiments, the accuracy obtained by the proposed approach (measured using Structural Hamming Distance) was found to be significantly better than traditional methods. Furthermore, the BN structure learned was found to be robust to variations in parameters, such as mean time to failure (MTTF).

Real-Time Sensing of Output Polymer Flow Temperature and Volumetric Flowrate in Fused Filament Fabrication Process.

In this paper we addressed key challenges in engineering an instrumentation system for sensing and signal processing for real-time estimation of two main process variables in the Fused-Filament-Fabrication process: (i) temperature of the polymer melt exiting the nozzle using a thermocouple; and (ii) polymer flowrate using extrusion width measurements in real-time, in-situ, using a microscope camera. We used a design of experiments approach to develop response surface models for two materials that enable accurate estimation of the polymer exit temperature as a function of polymer flowrate and liquefier temperature with a fit of R2=99.96% and 99.39%. The live video stream of the deposition process was used to compute the flowrate based on a road geometry model. Specifically, a robust extrusion width recognizer REXR algorithm was developed to identify edges of the deposited road and for real-time computation of extrusion width, which was found to be robust to filament colors and materials. The extrusion width measurement was found to be within 0.08 mm of caliper measurements with an R2 value of 99.91% and was found to closely track the requested flowrate from the slicer. This opens new avenues for advancing the engineering science for process monitoring and control of FFF.

Fusion-Learning of Bayesian Network Models for Fault Diagnostics.

Bayesian Network (BN) models are being successfully applied to improve fault diagnosis, which in turn can improve equipment uptime and customer service. Most of these BN models are essentially trained using quantitative data obtained from sensors. However, sensors may not be able to cover all faults and therefore such BN models would be incomplete. Furthermore, many systems have maintenance logs that can serve as qualitative data, potentially containing historic causation information in unstructured natural language replete with technical terms. The motivation of this paper is to leverage all of the data available to improve BN learning. Specifically, we propose a method for fusion-learning of BNs: for quantitative data obtained from sensors, metrology data and qualitative data from maintenance logs, corrective and preventive action reports, and then follow by fusing these two BNs. Furthermore, we propose a human-in-the-loop approach for expert knowledge elicitation of the BN structure aided by logged natural language data instead of relying exclusively on their anecdotal memory. The resulting fused BN model can be expected to provide improved diagnostics as it has a wider fault coverage than the individual BNs. We demonstrate the efficacy of our proposed method using real world data from uninterruptible power supply (UPS) fault diagnostics.

Sensor Selection Framework for Designing Fault Diagnostics System.

In a world of rapidly changing technologies, reliance on complex engineered systems has become substantial. Interactions associated with such systems as well as associated manufacturing processes also continue to evolve and grow in complexity. Consider how the complexity of manufacturing processes makes engineered systems vulnerable to cascading and escalating failures; truly a highly complex and evolving system of systems. Maintaining quality and reliability requires considerations during product development, manufacturing processes, and more. Monitoring the health of the complex system while in operation/use is imperative. These considerations have compelled designers to explore fault-mechanism models and to develop corresponding countermeasures. Increasingly, there has been a reliance on embedded sensors to aid in prognosticating failures, to reduce downtime, during manufacture and system operation. However, the accuracy of estimating the remaining useful life of the system is highly dependent on the quality of the data obtained. This can be enhanced by increasing the number of sensors used, according to information theory. However, adding sensors increases total costs with the cost of the sensors and the costs associated with information-gathering procedures. Determining the optimal number of sensors, associated operating and data acquisition costs, and sensor-configuration are nontrivial. It is also imperative to avoid redundant information due to the presence of additional sensors and the efficient display of information to the decision-maker. Therefore, it is necessary to select a subset of sensors that not only reduce the cost but are also informative. While progress has been made in the sensor selection process, it is limited to either the type of the sensor, number of sensors or both. Such approaches do not address specifications of the required sensors which are integral to the sensor selection process. This paper addresses these shortcomings through a new method, OFCCaTS, to avoid the increased cost associated with health monitoring and to improve its accuracy. The proposed method utilizes a scalable multi-objective framework for sensor selection to maximize fault detection rate while minimizing the total cost of sensors. A wind turbine gearbox is considered to demonstrate the efficacy of the proposed framework.

An Investigation to Manufacturing Analytical Services Composition using the Analytical Target Cascading Method.

As cloud computing is increasingly adopted, the trend is to offer software functions as modular services and compose them into larger, more meaningful ones. The trend is attractive to analytical problems in the manufacturing system design and performance improvement domain because 1) finding a global optimization for the system is a complex problem; and 2) sub-problems are typically compartmentalized by the organizational structure. However, solving sub-problems by independent services can result in a sub-optimal solution at the system level. This paper investigates the technique called Analytical Target Cascading (ATC) to coordinate the optimization of loosely-coupled sub-problems, each may be modularly formulated by differing departments and be solved by modular analytical services. The result demonstrates that ATC is a promising method in that it offers system-level optimal solutions that can scale up by exploiting distributed and modular executions while allowing easier management of the problem formulation.

Measuring agreement between healthcare survey instruments using mutual information.

BACKGROUND: Healthcare researchers often use multiple healthcare survey instruments to examine a particular patient symptom. The use of multiple instruments can pose some interesting research questions, such as whether the outcomes produced by the different instruments are in agreement. We tackle this problem using information theory, focusing on mutual information to compare outcomes from multiple healthcare survey instruments. METHODS: We review existing methods of measuring agreement/disagreement between the instruments and suggest a procedure that utilizes mutual information to quantitatively measure the amount of information shared by outcomes from multiple healthcare survey instruments. We also include worked examples to explain the approach. RESULTS: As a case study, we employ the suggested procedure to analyze multiple healthcare survey instruments used for detecting delirium superimposed on dementia (DSD) in community-dwelling older adults. In addition, several examples are used to assess the mutual information technique in comparison with other measures, such as odds ratio and Cohen's kappa. CONCLUSIONS: Analysis of mutual information can be useful in explaining agreement/disagreement between multiple instruments. The suggested approach provides new insights into and potential improvements for the application of healthcare survey instruments.

Modelling Adherence Behaviour for the Treatment of Obstructive Sleep Apnoea.

Continuous positive airway pressure therapy (CPAP) is known to be the most efficacious treatment for obstructive sleep apnoea (OSA). Unfortunately, poor adherence behaviour in using CPAP reduces its effectiveness and thereby also limits beneficial outcomes. In this paper, we model the dynamics and patterns of patient adherence behaviour as a basis for designing effective and economical interventions. Specifically, we define patient CPAP usage behaviour as a state and develop Markov models for diverse patient cohorts in order to examine the stochastic dynamics of CPAP usage behaviours. We also examine the impact of behavioural intervention scenarios using a Markov decision process (MDP), and suggest a guideline for designing interventions to improve CPAP adherence behaviour. Behavioural intervention policy that addresses economic aspects of treatment is imperative for translation to clinical practice, particularly in resource-constrained environments that are clinically engaged in the chronic care of OSA.