Integrating Artificial Intelligence (AI) with the Industrial Internet of Things (IIoT) has transformed industrial processes, enhancing productivity, quality control, and operational efficiency. However, ensuring the precision and reliability of sensor-generated data remains a critical challenge due to the evolving nature of industrial processes and the limitations of conventional validation methods. Traditional rule-based and supervised learning approaches struggle to adapt to process shifts, drifts, and novel anomalies, making sensor data validation an ongoing issue. This article introduces UDAVA (Unsupervised Learning Approach using Process Mining for Sensor Data Validation in IIoT), a novel AI-driven pipeline designed to automate the identification of reference patterns in sensor data and validate subsequent production cycles by recognizing deviations from expected behaviors. UDAVA employs a multi-stage process that includes preprocessing sensor data, clustering recurring patterns, and assessing deviations. It supports semi-supervised learning by integrating manual labels where available, improving interpretability and accuracy. One of UDAVA’s key strengths lies in its ability to extract features from sensor data rather than relying on raw time series similarity, making it robust against noise and diverse process variations. Additionally, UDAVA integrates process mining techniques—process discovery and conformance checking—to enhance its ability to detect even subtle anomalies and deviations in industrial workflows. We conduct a comprehensive evaluation of UDAVA using three industrial datasets, demonstrating its effectiveness in identifying high-level process behaviors, detecting process shifts and drifts, and ensuring data validation across multiple production cycles. The results highlight UDAVA ’s adaptability across different industrial processes, making it a valuable tool for optimizing operations and ensuring sensor data reliability in IIoT environments.

Deploying Large Language Models (LLMs) on edge devices presents significant challenges due to computational constraints, memory limitations, inference speed, and energy consumption. Model quantization has emerged as a key technique to enable efficient LLM inference by reducing model size and computational overhead. In this study, we conduct a comprehensive analysis of 28 quantized LLMs from the Ollama library, which applies by default Post-Training Quantization (PTQ) and weight-only quantization techniques, deployed on an edge device (Raspberry Pi 4 with 4 GB RAM). We evaluate energy efficiency, inference performance, and output accuracy across multiple quantization levels and task types. Models are benchmarked on five standardized datasets (CommonsenseQA, BIG-Bench Hard, TruthfulQA, GSM8K, and HumanEval), and we employ a high-resolution, hardware-based energy measurement tool to capture real-world power consumption. Our findings reveal the trade-offs between energy efficiency, inference speed, and accuracy in different quantization settings, highlighting configurations that optimize LLM deployment for resource-constrained environments. By integrating hardware-level energy profiling with LLM benchmarking, this study provides actionable insights for sustainable AI, bridging a critical gap in existing research on energy-aware LLM deployment.

The proliferation of machine learning (ML) has brought unprecedented advancements in technology, but it has also raised concerns about its environmental impact, particularly concerning carbon emissions. To address the imperative of environmentally responsible ML, we present in this paper a novel ML pipeline, named CEMAI, designed to monitor and analyze carbon emissions across the entire lifecycle of ML model development, from data preparation to training and deployment. Our endeavor involves an exhaustive evaluation process underpinned by three industrial case studies. These case studies are structured around the application of ML models to predict tool wear, estimate remaining useful lifetimes, and detect anomalies in the Industrial Internet of Things (IIoT). Leveraging sensor data originating from CNC machining and broaching operations, our research shows empirically the efficacy of carbon emissions as a dependable metric guiding the configuration of an ML development process. The essence of our approach lies in striking a balance between superior performance and minimal carbon emissions. Our findings reveal the potential to optimize pipeline configurations for ML models in a manner that not only enhances performance but also drastically reduces carbon emissions, thereby underlining the significance of adopting ecologically responsible engineering practices.

Abstract Workers, especially those involved in jobs requiring extended standing or repetitive movements, often face significant health challenges due to Musculoskeletal Disorders (MSDs). To mitigate MSD risks, enhancing workplace ergonomics is vital, which includes forecasting long-term employee postures, educating workers about related occupational health risks, and offering relevant recommendations. However, research gaps remain, such as the lack of a sustainable AI/ML pipeline that combines sensor-based, uncertainty-aware posture prediction with large language models for natural language communication of occupational health risks and recommendations. We introduce ERG-AI, a machine learning pipeline designed to predict extended worker postures using data from multiple wearable sensors. Alongside providing posture prediction and uncertainty estimates, ERG-AI also provides personalized health risk assessments and recommendations by generating prompts based on its performance and prompting Large Language Model (LLM) APIs, like GPT-4, to obtain user-friendly output. We used the Digital Worker Goldicare dataset to assess ERG-AI, which includes data from 114 home care workers who wore five tri-axial accelerometers in various bodily positions for a cumulative 2913 hours. The evaluation focused on the quality of posture prediction under uncertainty, energy consumption and carbon footprint of ERG-AI and the effectiveness of personalized recommendations rendered in easy-to-understand language.

The Industrial Internet of Things (IIoT) leverages thousands of interconnected sensors and computing devices to monitor and control large and complex industrial processes. Machine learning (ML) applications in IIoT use data acquired from multiple sensors to perform tasks such as predictive maintenance. While remembering useful learning from the past, these applications need to adapt learning for evolving sensor data stemming from changes in industrial processes and environmental conditions. This paper presents a continual learning pipeline to learn from the evolving data while replaying selected parts of the old data. The pipeline is configured to produce ML experiences (e.g., training a baseline neural network model), improve the baseline model with the new data while replaying part of the old data, and infer/predict using a specific model version given a stream of IIoT sensor data. We have evaluated our approach from an AI Engineering perspective using three industrial case studies, i.e., predicting tool wear, remaining useful lifetime, and anomalies from sensor data acquired from CNC machining and broaching operations. Our results show that configuring experiences for replay-driven continual learning allows dynamic maintenance of ML performance on evolving data while minimizing the excessive accumulation of legacy sensor data.

Unsupervised learning systems using clustering have gained significant attention for numerous applications due to their unique ability to discover patterns and structures in large unlabeled datasets. However, their effectiveness highly depends on their configuration, which requires domain-specific expertise and often involves numerous manual trials. Specifically, selecting appropriate algorithms and hyperparameters adds to the complexity of the configuration process. In this paper, we propose, apply, and assess an automated approach (AutoConf) for configuring unsupervised learning systems using clustering, leveraging metamorphic testing and Bayesian optimization. Metamorphic testing is utilized to verify the configurations of unsupervised learning systems by applying a series of input transformations. We use Bayesian optimization guided by metamorphic-testing output to automatically identify the optimal configuration. The approach aims to streamline the configuration process and enhance the effectiveness of unsupervised learning systems. It has been evaluated through experiments on six datasets from three domains for anomaly detection. The evaluation results show that our approach can find configurations outperforming the baseline approaches as they achieved a recall of 0.89 and a precision of 0.84 (on average).

Manufacturing has enabled the mechanized mass production of the same (or similar) products by replacing craftsmen with assembly lines of machines. The quality of each product in an assembly line greatly hinges on continual observation and error compensation during machining using sensors that measure quantities such as position and torque of a cutting tool and vibrations due to possible imperfections in the cutting tool and raw material. Patterns observed in sensor data from a (near-)optimal production cycle should ideally recur in subsequent production cycles with minimal deviation. Manually labeling and comparing such patterns is an insurmountable task due to the massive amount of streaming data that can be generated from a production process. We present UDAVA, an unsupervised machine learning pipeline that automatically discovers process behavior patterns in sensor data for a reference production cycle. UDAVA performs clustering of reduced dimensionality summary statistics of raw sensor data to enable high-speed clustering of dense time-series data. It deploys the model as a service to verify batch data from subsequent production cycles to detect recurring behavior patterns and quantify deviation from the reference behavior. We have evaluated UDAVA from an AI Engineering perspective using two industrial case studies.

Increasing the impact of software engineering research in the software industry and the society at large has long been a concern of high priority for the software engineering community. The problem of two cultures, research conducted in a vacuum (disconnected from the real world), or misaligned time horizons are just some of the many complex challenges standing in the way of successful industry–academia collaborations. This article reports on the experience of research collaboration and knowledge co-creation between industry and academia in software engineering as a way to bridge the research–practice collaboration gap. Our experience spans 14 years of collaboration between researchers in software engineering and the European and Norwegian software and IT industry. Using the participant observation and interview methods, we have collected and afterwards analyzed an extensive record of qualitative data. Drawing upon the findings made and the experience gained, we provide a set of 14 patterns and 14 anti-patterns for industry–academia collaborations, aimed to support other researchers and practitioners in establishing and running research collaboration projects in software engineering.

Effective assessment of cyber risks in the increasingly dynamic threat landscape must be supported by artificial intelligence techniques due to their ability to dynamically scale and adapt. This article provides the state of the art of AI-supported security risk assessment approaches in terms of a systematic mapping study. The overall goal is to obtain an overview of security risk assessment approaches that use AI techniques to identify, estimate, and/or evaluate cyber risks. We carried out the systematic mapping study following standard processes and identified in total 33 relevant primary studies that we included in our mapping study. The results of our study show that on average, the number of papers about AI-supported security risk assessment has been increasing since 2010 with the growth rate of 133% between 2010 and 2020. The risk assessment approaches reported have mainly been used to assess cyber risks related to intrusion detection, malware detection, and industrial systems. The approaches focus mostly on identifying and/or estimating security risks, and primarily make use of Bayesian networks and neural networks as supporting AI methods/techniques.

Obstructive sleep apnea is a serious sleep disorder that affects an estimated one billion adults worldwide. It causes breathing to repeatedly stop and start during sleep which over years increases the risk of hypertension, heart disease, stroke, Alzheimer's, and cancer. In this demo, we present Yolo4Apnea a deep learning system extending You Only Look Once (Yolo) system to detect sleep apnea events from abdominal breathing patterns in real-time enabling immediate awareness and action. Abdominal breathing is measured using a respiratory inductance plethysmography sensor worn around the stomach. The source code is available at https://github.com/simula-vias/Yolo4Apnea

Tracking physical effort from physiological signals has enabled people to manage required activity levels in our increasingly sedentary and automated world. Breathing is a physiological process that is a reactive representation of our physical effort. In this demo, we present DeepVentilation, a deep learning system to predict minute ventilation in litres of air a person moves in one minute uniquely from real-time measurement of rib-cage breathing forces. DeepVentilation has been trained on input signals of expansion and contraction of the rib-cage obtained using a non-invasive respiratory inductance plethysmography sensor to predict minute ventilation as observed from a face/head mounted exercise spirometer. The system is used to track physical effort closely matching our perception of actual exercise intensity. The source code for the demo is available here: https://github.com/simula-vias/DeepVentilation

Requirements Engineering (RE) has established itself as a software engineering discipline over the past decades. While researchers have been investigating the RE discipline with a plethora of empirical studies, attempts to systematically derive an empirical theory in context of the RE discipline have just recently been started. However, such a theory is needed if we are to define and motivate guidance in performing high quality RE research and practice. We aim at providing an empirical and externally valid foundation for a theory of RE practice, which helps software engineers establish effective and efficient RE processes in a problem-driven manner. We designed a survey instrument and an engineer-focused theory that was first piloted in Germany and, after making substantial modifications, has now been replicated in 10 countries worldwide. We have a theory in the form of a set of propositions inferred from our experiences and available studies, as well as the results from our pilot study in Germany. We evaluate the propositions with bootstrapped confidence intervals and derive potential explanations for the propositions. In this article, we report on the design of the family of surveys, its underlying theory, and the full results obtained from the replication studies conducted in 10 countries with participants from 228 organisations. Our results represent a substantial step forward towards developing an empirical theory of RE practice. The results reveal, for example, that there are no strong differences between organisations in different countries and regions, that interviews, facilitated meetings and prototyping are the most used elicitation techniques, that requirements are often documented textually, that traces between requirements and code or design documents are common, that requirements specifications themselves are rarely changed and that requirements engineering (process) improvement endeavours are mostly internally driven. Our study establishes a theory that can be used as starting point for many further studies for more detailed investigations. Practitioners can use the results as theory-supported guidance on selecting suitable RE methods and techniques.

DevOps, as a growing development practice that aims to enable faster development and efficient deployment of applications without compromising on quality, is often hampered by long cycle times. One contributing factor to long cycle times in DevOps is long build time. Automated testing in continuous integration is one of the build stages that is highly prone to long run-time due to software complexity and evolution, and inefficient due to unoptimized testing approaches. To be cost-effective, testing in continuous integration needs to use only a fast-running set of comprehensive tests that are able to ensure the level of quality needed for deployment to production. Known approaches use time-aware test selection methods to improve time-efficiency of continuous integration testing by providing optimized combinations and order of tests with respect to decreased run-time. However, focusing on time-efficiency as the sole criterion in DevOps often jeopardizes the quality of software deliveries. This paper proposes a technique that integrates fault-based and risk-based test selection and prioritization optimized for low run-time, to improve time-effectiveness of continuous integration testing, and thus reduce long cycle times in DevOps, without compromising on quality. The technique has been evaluated in testing of a large-scale configurable software in continuous integration, and has shown considerable improvement over industry practice with respect to time-efficiency.

Growing evidence suggests the DevOps approach enables faster development and deployment, and easier maintenance of applications.Still, the efficiency of DevOps is constrained by long cycle times.This paper presents the approach for improving time-efficiency in DevOps, and in particular continuous integration testing, using continuous test optimization.The approach uses test redundancy analysis to discover test overlap with respect to feature interaction coverage, and based on detected redundancy to reduce the size of a test suite.Smallersize test suites execute faster and enable shorter test cycles, which further enables shorter release cycles.The approach has been experimentally evaluated using an industrial case study, against three metrics: industry practice of test selection for continuous integration testing, retest-all approach, and random test selection.The results suggest that the proposed test redundancy detection and reduction efficiently reduces test cycles in CI compared to industry practice and retest-all approach, and improves faultdetection effectiveness compared to random test selection 1 .

Elasticity is one of the main features of cloud-based systems (CBSs), where elastic adaptations, such as those to deal with scaling in or scaling out of computational resources, help meet performance requirements under varying workload. There is an industrial need to find configurations of elastic adaptations and workload that could lead to degradation of performance in a CBS, serving possibly millions of users. However, the potentially great number of such configurations poses a challenge: executing and verifying all of them on the cloud can be prohibitively expensive in both, time and cost. We present an approach to model elasticity adaptation due to workload changes as a classification tree model and consequently generate short test sequences of configurations that cover all T-wise interactions between parameters in the model. These test sequences, when executed, help us assess the performance of elastic CBS. Using MongoDB as a case study, test sequences generated by our approach reveal several significant performance degradations.

Exhaustive testing of highly configurable software developed in continuous integration is rarely feasible in practice due to the configuration space of exponential size on the one hand, and strict time constraints on the other. This entails using selective testing techniques to determine the most failure-inducing test cases, conforming to highly-constrained time budget. These challenges have been well recognized by researchers, such that many different techniques have been proposed. In practice, however, there is a lack of efficient tools able to reduce high testing effort, without compromising software quality. In this paper we propose a test suite optimization technology TITAN, which increases the time-and cost-efficiency of testing highly configurable software developed in continuous integration. The technology implements practical test prioritization and minimization techniques, and provides test traceability and visualization for improving the quality of testing. We present the TITAN tool and discuss a set of methodological and technological challenges we have faced during TITAN development. We evaluate TITAN in testing of Cisco's highly configurable software with frequent high quality releases, and demonstrate the benefit of the approach in such a complex industry domain.

Data-intensive systems in e-governance collect and process data to ensure conformance to a set of business rules. Testers meticulously verify data in test databases, extracted from different steps of a live production stream , for correct application of business rules. We simplify the process by allowing testers to model a test domain on a relational database and automatically generate test cases representing data interactions satisfying combinatorial interaction coverage criteria. This paper also introduces test cases with self-referential interactions, which is a necessity in real-world databases. We verify these test cases using our human-in-the-loop tool, Depict. Depict, with expert assistance, generates complex SQL queries for test cases and produces a visual report of test case satisfaction. We apply the approach to two scenarios: 1) simplify and optimize a periodic archiving operation and 2) verify fault codes within the testing environment of the Custom directorate's TVINN system.

Differing perceptions and expectations are obstaclesto collaboration between software engineering (SE) researchersand practitioners: Researchers often have a view thatpractitioners are reluctant to share real data. Practitionersbelieve that researchers are mostly working on topics which aredivorced from real industrial needs. Researchers believe thatpractitioners are looking for quick fixes. Practitioners have aview that case studies in research do not represent thecomplexities of real projects. Researchers may expect a few yearsto do research on a problem whereas practitioners expect a quicksolution that pays off immediately.Researchers and practitioners need to identify the gaps and todiscover the ways to collaborate to strengthen SE research andindustrial practice (IP). The main purpose of this workshop is tobring together researchers and practitioners to discuss thecurrent state of SE research and IP and to enhance collaborationbetween them. The SER&IP 2015 workshop provided a platformto share success stories of SE research-practice partnerships aswell as to discuss the challenges, through a day-long agenda ofkeynotes, paper presentations and round table discussions.

Specifying a model transformation is challenging as it must be able to give a meaningful output for any input model in a possibly infinite modeling domain. Transformation pre-conditions constrain the input domain by rejecting input models that are not meant to be transformed by a model transformation. This paper presents a systematic approach to discover such pre-conditions when it is hard for a human developer to foresee complex graphs of objects that are not meant to be transformed. The approach is based on systematically generating a finite number of test models using our tool, PRAMANA to first cover the input domain based on input domain partitioning. Tracing a transformation's execution reveals why some pre-conditions are missing. Using a benchmark transformation from simplified UML class diagram models to RDBMS models we discover new pre-conditions that were not initially specified.

Testing data-intensive systems is paramount to increase our reliance on information processed in e-governance, scientific/ medical research, and social networks. A common practice in the industrial testing process is to use test databases copied from live production streams to test functionality of complex database applications that manage well-formedness of data and its adherence to business rules in these systems. This practice is often based on the assumption that the test database adequately covers realistic scenarios to test, hopefully, all functionality in these applications. There is a need to systematically evaluate this assumption. We present a tool-supported method to model realistic scenarios and verify whether copied test databases actually cover them and consequently facilitate adequate testing. We conceptualize realistic scenarios as data interactions between fields cross-cutting a complex database schema and model them as test cases in a classification tree model. We present a human-in the-loop tool, DEPICT, that uses the classification tree model as input to (a) facilitate interactive selection of a connected sub graph from often many possible paths of interactions between tables specified in the model (b) automatically generate SQL queries to create an inner join between tables in the connected sub graph (c) extract records from the join and generate a visual report of satisfied and unsatisfied interactions hence quantifying test adequacy of the test database. We report our experience as a qualitative evaluation of approach and with a large industrial database from the Norwegian Customs and Excise information system TVINN featuring large and complex databases with millions of records.

Regression testing in continuous integration environment is bounded by tight time constraints. To satisfy time constraints and achieve testing goals, test cases must be efficiently ordered in execution. Prioritization techniques are commonly used to order test cases to reflect their importance according to one or more criteria. Reduced time to test or high fault detection rate are such important criteria. In this paper, we present a case study of a test prioritization approach ROCKET (Prioritization for Continuous Regression Testing) to improve the efficiency of continuous regression testing of industrial video conferencing software. ROCKET orders test cases based on historical failure data, test execution time and domain-specific heuristics. It uses a weighted function to compute test priority. The weights are higher if tests uncover regression faults in recent iterations of software testing and reduce time to detection of faults. The results of the study show that the test cases prioritized using ROCKET (1) provide faster fault detection, and (2) increase regression fault detection rate, revealing 30% more faults for 20% of the test suite executed, comparing to manually prioritized test cases.

Traceability between development artefacts and mainly from and to requirements plays a major role in system lifecycle, supporting activities such as system validation, change impact analysis, and regulation compliance. Many researchers have been working on this topic and have published their work throughout the editions of the Requirements Engineering Conference. This paper aims to analyse the research on traceability published in the past 20 years of this conference and to provide insights into its contribution to the traceability area. We have selected and reviewed 70 papers in the proceedings of the conference and summarised several aspects of traceability that have been addressed and by whom. The paper also discusses the evolution of the topic at the conference, compares the results with those reported in other publications, and proposes aspects on which further research should be conducted.

One key challenge for software product lines is efficiently managing variability throughout their lifecycle. In this paper, we address the problem of variability in software product lines testing. We (1) identify a set of issues that must be addressed to make software product line testing work in practice and (2) provide a framework that combines a set of techniques to solve these issues. The framework integrates feature modelling, combinatorial interaction testing and constraint programming techniques. First, we extract variability in a software product line as a feature model with specified feature interdependencies. We then employ an algorithm that generates a minimal set of valid test cases covering all 2-way feature interactions for a given time interval. Furthermore, we evaluate the framework on an industrial SPL and show that using the framework saves time and provides better test coverage. In particular, our experiments show that the framework improves industrial testing practice in terms of (i) 17% smaller set of test cases that are (a) valid and (b) guarantee all 2-way feature coverage (as opposite to 19.2% 2-way feature coverage in the hand made test set), and (ii) full flexibility and adjustment of test generation to available testing time.

Model transformations are an integral part of several computing systems that manipulate interconnected graphs of objects called models in an input domain specified by a metamodel and a set of invariants. Test models are used to look for faults in a transformation. A test model contains a specific set of objects, their interconnections and values for their attributes. Can we automatically generate an effective set of test models using knowledge from the transformation? We present a white-box testing approach that uses static analysis to guide the automatic generation of test inputs for transformations. Our static analysis uncovers knowledge about how the input model elements are accessed by transformation operations. This information is called the input metamodel footprint due to the transformation. We transform footprint, input metamodel, its invariants, and transformation pre-conditions to a constraint satisfaction problem in Alloy. We solve the problem to generate sets of test models containing traces of the footprint. Are these test models effective? With the help of a case study transformation we evaluate the effectiveness of these test inputs. We use mutation analysis to show that the test models generated from footprints are more effective (97.62% avg. mutation score) in detecting faults than previously developed approaches based on input domain coverage criteria (89.9% avg.) and unguided generation (70.1% avg.).

Software Product Lines (SPL) are difficult to validate due to combinatorics induced by variability across their features. This leads to combinatorial explosion of the number of derivable products. Exhaustive testing in such a large space of products is infeasible. One possible option is to test SPLs by generating test cases that cover all possible T feature interactions (T-wise). T-wise dramatically reduces the number of test products while ensuring reasonable SPL coverage. However, automatic generation of test cases satisfying T-wise using SAT solvers raises two issues. The encoding of SPL models and T-wise criteria into a set of formulas acceptable by the solver and their satisfaction which fails when processed “all-at-once'”. We propose a scalable toolset using Alloy to automatically generate test cases satisfying T-wise from SPL models. We define strategies to split T-wise combinations into solvable subsets. We design and compute metrics to evaluate strategies on Aspect OPTIMA, a concrete transactional SPL.

The ever-growing choice in diverse services is making service orchestration variability an essential aspect of a composite web service. Influence of this variation on the Quality of Service (QoS) of a composite service is critical and the focus of our work. In this paper, we present a methodology to first model orchestration variability using a feature diagram (FD). The FD specifies a product line of orchestrations represented as configurations of invoked/rejected atomic services. Second, due to the potentially large set of configurations we employ combinatorial testing techniques to automatically generate configurations covering all valid pair wise interactions between services. Third, we analyze QoS variation for each configuration using probabilistic models of QoS. Using a crisis management system case study we experimentally show that pair wise generation covers all QoS outliers and eliminates analysis of > 75% of all possible configurations. The QoS analysis of the pair wise configurations reveals unsafe/ineffective configurations, helps determine realistic Service Level Agreements (SLAs), and provides valuable feedback to help remodel an orchestration.

Testing remains a major challenge for model transformation development. Test models that are used as test data for model transformations, are constrained by various sources of knowledge that is expressed in different formalisms. Thus, in order to automatically generate test models it is necessary to interpret these different sources of knowledge and combine them into a consistent set of information that can be used for model synthesis. In this paper, we identify sources of testing knowledge and present our tool Cartier that uses Alloy as the first-order relational logic language to represent combined knowledge in the form of constraints. The constraints are solved leading to a selection of qualified test models from the input domain of a model transformation. We illustrate our approach using the Unified Modeling Language class diagram to relational database management systems transformation as a running example.