Fault-Tolerant Humidity Sensor Based on Neural Networks

The article discusses fault-tolerant neural network moisture sensors development methodology for pasty and bulk materials. The input of the neural network humidity sensor receives information from primary measuring transducers that control materials' drying process in real time inside various drying installations. The relative moisture value obtained from the sensor output. Proposed methodology is based on primary measuring transducers failure situations classification. The following failure classes are introduced: failure-free operation of all primary measuring transducers, soft failure, medium failure, hard failure. Depending on the class, models are created to estimate materials relative moisture by training neural networks with failures. Neural-based moisture sensors for bulk and pasty materials development examples are given. Proposed fault-tolerant neural network humidity sensors are already used in information and control systems for pasty materials control. Based on the information received, a decision is made to control the drying installation to minimize losses in drying process quality and productivity. Industrial complexes modernization through intelligent information and control systems introduction offers a significant products competitiveness increase manufactured by domestic enterprises.