Quality of product is an important aspect in many commercial organizations where storage and shipment practices are required. Temperature is one of the main parameters that influence quality and temperature treatments of agricultural products therefore require special attention.

The temperature variation in a meat chiller has a significant effect on tenderness, color and microbial status of the meat, therefore thermal mapping during the chilling process and during chilled shipment to overseas markets is vital. The literature indicates that deviations of only a few degrees can lead to significant product deterioration. There are several existing methods for thermal mapping: these includes Computational Fluid Dynamics (CFD), Finite Element Methods (FEM) for examination of the environmental variables in the chiller. These methodologies can work effectively in non real-time. However these methods are quite complex and need high computational overhead when it comes to hard real-time analysis within the context of the process dynamics.

The focus of this research work is to develop a method and system towards building an object-centric environment monitoring using collaborative efforts of both wireless sensor networks and artificial neural networks for spatial thermal mapping. Thermal tracking of an object placed anywhere within a predefined space is one of the main objectives here. Sensing data is gathered from restricted sensing points and used for training the Neural Network on the spatial distribution of the temperature at a given time. The solution is based on the development of a generic module that could be used as a basic building block for larger spaces. The Artificial Neural Networks (ANNs) perform dynamic learning using the data it collects from the various sensing points within the specific subspace module. The ANN could then be used to facilitate mapping of any other point in the related sub-space. The distribution of the sensors (nodes placement strategy for better coverage) is used as a parameter for evaluating the ability to predict the temperature at any point within the space.

This research work exploits the neuro Wireless Sensor Network (nWSN) architecture in steady-state and transient environments. A conceptual model has been designed and built in a simulation environment and also experiments conducted using a test-bed. A Shepard’s algorithm with modified Euclidian distance is used for comparison with an adaptive neural network solution. An algorithm is developed to divide the overall space into subspaces covered by clusters of neighbouring sensing nodes to identify the thermal profiles. Using this approach, a buffering and Query based nWSN Data Processing (QnDP) algorithm is proposed to fulfil the data synchronization. A case study on the meat plants cool storage has been undertaken to demonstrate the best layout and location identification of the sensing nodes that can be attached to the carcasses to record thermal behavior.

This research work assessed the viability of using nWSN architecture. It found that the Mean Absolute Error (MAE) at the infrastructural nodes has a variation of less than 0.5C. The resulting MAE is effective when nWSN can be capable of generating similar applications of predictions.