This study aims to identify an effective neural network architecture for the task of semantic segmentation of the surface of beer wort at the stage of primary fermentation, using deep learning methodologies. Four contemporary architectures were evaluated and contrasted. The following networks are presented in both baseline and optimized forms: U-Net, DeepLabV3+, Fast-SCNN, and SegNet. The models were trained on a dataset of images depicting real beer surfaces at the primary fermentation stage. This was followed by the validation of the models using key metrics, including pixel classification accuracy, Mean Intersection over Union (mIoU), Dice Coefficient, inference time per image, and Graphics Processing Unit (GPU) resource utilization. Results indicate that the optimized U-Net achieved the optimal balance between performance and efficiency, attaining a validation accuracy of 88.85%, mIoU of 76.72%, and a Dice score of 86.71%. With an inference time of 49.5 milliseconds per image, coupled with minimal GPU utilization (18%), the model proves suitable for real-time deployment in production environments. Conversely, complex architectures, such as DeepLabV3+, did not yield the anticipated benefits, thereby underscoring the viability of utilizing compact models for highly specialized industrial tasks. This study establishes a novel quantitative metric for the assessment of fermentation. This is based on the characteristics of the foam surface and thus offers an objective alternative to traditional subjective inspections. The findings emphasize the potential of adapting optimized deep learning architectures to quality control tasks within the food industry, particularly in the brewing sector, and they pave the way for further integration into automated computer vision systems.

Reliable schedule-risk estimation in hybrid software development lifecycles is strategically important for organizations adopting AI in software engineering. This study addresses that need by transforming routine process telemetry (CI/CD, SAST, traceability) into explainable, quantitative predictions of completion time and rework. This paper introduces an integrated probabilistic model of the hybrid software development lifecycle that combines Generalized Evaluation and Review Technique (GERT) network semantics with I-AND synchronization, explicit artificial-intelligence (AI) interventions, and a fuzzy treatment of epistemic uncertainty. The model embeds two controllable AI nodes–an AI Requirements Assistant and AI-augmented static code analysis, directly into the process topology and applies an analytical reduction to a W-function to obtain iteration-time distributions and release-success probabilities without resorting solely to simulation. Epistemic uncertainty on critical arcs is represented by fuzzy intervals and propagated via Zadeh’s extension principle, while aleatory variability is captured through stochastic branching. Parameter calibration relies on process telemetry (requirements traceability, static-analysis signals, continuous integration/continuous delivery, CI/CD, and history). A validation case (“system design → UX prototyping → implementation → quality assurance → deployment”) demonstrates practical use: large samples of process trajectories are generated under identical initial conditions and fixed random seeds, and kernel density estimation with Silverman’s bandwidth is applied to normalized histograms of continuous outcomes. Results indicate earlier defect detection, fewer late rework loops, thinner right tails of global duration, and an approximately threefold reduction in the expected number of rework cycles when AI is enabled. The framework yields interpretable, scenario-ready metrics for tuning quality-gate policies and automation levels in Agile/DevOps settings.

This paper presents a prototype of an autonomous hydrochemical monitoring system developed for large freshwater aquaculture facilities, directly addressing the need for smart monitoring in Agriculture 4.0. The proposed solution employs low-power sensor nodes based on commercially available components and long-range LoRaWAN communication to achieve continuous, scalable, and energy-efficient water quality monitoring. Each sensor module performs on-board signal preprocessing, including anomaly detection and short-term forecasting of key hydrochemical parameters. An ecological pond dynamics model incorporating an Extended Kalman Filter is used to fuse heterogeneous sensor data with predictive estimates, thus increasing measurement reliability. High-level data analysis, long-term storage, and cross-site comparison are performed on the server side. This integration enables adaptive tracking of environmental variations, supports early detection of hazardous trends associated with fish mortality risks, and allows one to explain and justify the reasoning behind every recommended corrective action. The performance of the forecasting and filtering algorithms is evaluated, and key system characteristics—including measurement accuracy, power consumption, and scalability—are discussed. Preliminary tests of the system prototype have shown that it can predict the dissolved oxygen level with $RMSE$ = 0.104 mg/L even with a minimum set of sensors. The results demonstrate that the proposed conceptual design of the system can be used as a base for real-time monitoring and predictive assessment of hydrochemical conditions in aquaculture environments.