Conference Agenda

Gas concentration sensors are increasingly adopted in intensive livestock systems to ensure indoor air quality and comply with regulations. However, their potential contribution to predicting on-farm electrical power demand remains unexplored. The objective of this study is to investigate how gas concentration data can support machine learning models for predicting farm power demand. Within the Biotegania project, a commercial farrowing room hosting 40 sows was monitored for six months. Three indoor measuring points were installed to monitor air temperature, relative humidity, and NH₃ and CO₂ concentrations, while an external point provided outdoor weather conditions. An IoT wireless clamp meter system recorded the electric current absorbed by heating, lighting, and ventilation systems, with a 10-minute timestep. Machine learning models were implemented to predict room power demand using the monitored indoor environmental conditions as independent variables. Preliminary experiments were performed using XGBoost on 13,125 data points for training and 3,304 for testing. The results show the algorithm achieved promising performances on test, with a WAPE of 13%. These findings suggest that monitoring gas concentrations in livestock facilities may support not only the estimation of gaseous emissions and the improvement of animal welfare, but also the real-time prediction of on-farm power demand.

Wood pellets are a popular source of biomass due to their high energy density and ease of storage compared to other solid biofuels. Limited mechanical durability, high sensitivity to moisture, and the tendency to generate fines during handling represent the main challenges associated with pellet handling, affecting structural stability, energy efficiency, and operational reliability.

This work investigates the effects of handling processes on densified wood pellets. An initial sample of bagged pine wood pellets (6 mm ± 0.1 mm, ENplus A1) was characterized for dimensions, moisture content, bulk density, and mechanical properties (cohesion and internal friction angle) under as-supplied conditions. The pellets were subjected to ten filling and discharge cycles in an experimental model silo with corrugated steel walls (0.152 m³). A post-handling sample was collected to evaluate changes in pellet dimensions, bulk density, and shear strength parameters.

Results reveal significant differences between fresh and handled pellets. Mean pellet length decreased from 14.62 mm to 12.69 mm, the proportion of smaller particles increased, and the internal friction angle rose from 41° to 52° after handling. These changes in physico-mechanical properties should be considered in the design of equipment for biomass storage and handling to ensure stability and operational reliability.

A silo model has been used to compare the influence of the wall type on wall pressures, mass flow rate, lateral pressure ratio, or effective wall friction coefficient. The silo model consists of a vertical section 0.75 m high, with a square cross-section of 0.45 x 0.45 m, and a flat bottom with a centric outlet measuring 0.06 x 0.06 m. Two different steel wall configurations were tested: a smooth sheet and a corrugated wall.

Normal and tangential pressures were recorded on both opposite silo walls in 0.15 x 0.15 m test panels.

The silo module is supported by four beam-type load cells, which measure the total wall friction forces generated by the stored bulk solid. In addition, four additional load cells support the flat bottom, providing the weight of bulk material resting over the silo bottom. Pinewood pellets were employed for conducting the tests.

Significant differences are observed in those parameters influenced by wall friction. Thus, the wall frictional pressures at the end of the filling are more than two times higher for the corrugated wall (375 Pa vs 153 Pa), as well as the effective friction coefficient (0.58 vs 0.24).

Drainage wells are designed to maintain lower groundwater levels and prevent landslides. Liner plates in steel drainage wells are subject to long-term progressive corrosion. Conventional visual inspection requires engineers to physically enter the well, while image processing applied to omnidirectional camera imagery has been reported to exhibit strong dependence on parameter settings and limited capacity to handle complex corrosion patterns. A quantitative corrosion detection method based on deep learning is examined in this study, applied to images of well interiors acquired with an omnidirectional camera. Classification tasks are defined: binary classification (corroded or non-corroded) and multi-class classification, including individual classes such as corrosion, vegetation, and reinforced materials. The investigated drainage wells are located in Tokamachi City, Niigata Prefecture, Japan. The detection performance is then compared between the two approaches. Thus, the multiclass classification model achieved an F1-score of 0.95, while the binary classification model yielded 0.94, confirming that the multiclass model provided superior corrosion detection performance. This performance difference is attributed to the heterogeneity of the non-corroded class in binary classification, where visually dissimilar elements, such as vegetation and repair materials, cause feature representations to scatter in the feature space, thereby reducing decision boundary performance.

Aging irrigation dams face unprecedented challenges from long-term deterioration and environmental factors. With agriculture accounting for approximately 70% of global freshwater withdrawals, maintaining structural integrity is critical for water security. This study develops a Physics-Guided Neural Network (PGNN) approach for high-accuracy surface temperature monitoring to detect subsurface damage in irrigation dams. The methodology integrates physics-based heat balance principles with neural network learning to analyze concrete surface temperature. The model evaluates net radiation, sensible heat transfer, and ground conduction heat while incorporating thermal properties of concrete as physical constraints. UAV-mounted thermal imaging and continuous thermocouple monitoring provide validation data. The PGNN model achieves superior accuracy (MAE = 0.9°C) compared to conventional heat balance analysis (MAE = 2.5°C) and standard LSTM (MAE = 1.0°C). By incorporating thermal transfer physics into the neural network architecture, the model accurately distinguishes thermal response patterns between damaged and non-damaged parts, enabling early detection before visible surface deterioration appears. This cost-effective, non-contact monitoring approach enables comprehensive assessment of large irrigation dam surfaces, supporting sustainable water infrastructure management and climate-resilient agricultural practices through predictive maintenance strategies.

Digital twins are emerging as a powerful tool for performance-driven design in agricultural buildings, yet their application to airflow-intensive systems remains limited. This study develops a BIM-driven workflow integrating lumped-parameter airflow modeling for the design of solar barn dryers used in forced-ventilation hay drying.

Nine alternative configurations were generated within a BIM environment to systematically explore the impact of airflow layout while maintaining constant processing capacity. The airflow network was modeled as a system of hydraulic resistances, capturing both distributed and concentrated losses.

Results demonstrate that airflow circuit geometry is a primary driver of system performance. Cumulative pressure losses vary by 10–20% across layouts, leading to differences of up to ~40% in required fan power (≈11–18 kW) and about one-third in annual electricity consumption. Configurations with smoother transitions and more compact, symmetric layouts consistently reduce energy demand while maintaining adequate overpressure.

These findings show that early-stage geometric decisions have a first-order impact on both energy use and system reliability. The proposed BIM–lumped parameter framework enables rapid, physically grounded design evaluation and supports the identification of cost-effective, energy-efficient solutions for solar barn dryers.

Precision livestock farming requires continuous monitoring and control of indoor environmental conditions to ensure animal welfare, optimize productivity, and reduce energy consumption. This study presents the implementation and one-year results of a smart IoT-based monitoring and control system deployed at a representative intensive swine farm in the Po Valley (Italy), focusing on nursery and growing-finishing buildings.

The system is based on a custom-designed RES4LIVE gateway (Raspberry Pi-based), integrating heterogeneous environmental and energy sensors through multiple communication protocols (Modbus RTU/TCP, M-Bus, MQTT, REST). Environmental monitoring included temperature, relative humidity, CO₂, NH₃, H₂S, O₂, and VOC sensors distributed in four zones across two buildings, alongside energy meters and a weather station. Data were locally stored and transmitted to a cloud platform via MQTT, enabling real-time visualization, rule-based smart control, health monitoring, and remote actuation. Advanced features include automated data validation, failure mitigation strategies, and demand-oriented energy management.

This paper presents the analysis of one year of thermo-hygrometric and air quality data, highlighting seasonal trends, critical conditions, and system reliability. The results demonstrate the robustness of the architecture and its potential to support precision environmental control in intensive pig production systems.

Detecting leakage in aging underground pipelines remains challenging due to limited accessibility and complex hydraulic conditions. This study investigates a transient damping-based leakage detection method applied to an existing pump-driven pipeline. Both internal water pressure and external hoop strain measurements were used. Field tests were conducted on a 5.5 km pressurized agricultural pipeline during pump-stopping operations. The first, third and fifth harmonic components of the pressure transient are extracted using band-pass filtering. Leakage-induced damping factors are then calculated for each component. Leakage locations are estimated by comparing the ratios of these damping factors with theoretical distributions. Pressure-based estimation yields errors of 6.1–13.3%. Strain-based estimation shows errors of 5.3–33.6%. Although hoop strain-based estimation exhibited lower accuracy, similar trends are observed in the damping characteristics. This demonstrates that external strain measurements can indirectly capture pressure transient behavior. Notably, this is one of the first experimental verifications of the transient damping-based method on a km-scale in-service pipeline. These results support the potential of this approach as a practical and non-intrusive diagnostic tool for aging infrastructure where reflection-based methods are ineffective.