A Sensor Location-based Diagnosis Method for Temperature Uniformity of Hotplate with Convolution Neural Network

A set of resistance temperature detectors (RTDs), which are temperature sensors that change resistance with temperature, is embedded in a hotplate to ensure uniform temperature control. Sensor failures can occur when the material of an RTD oxidizes due to thermal stress from the hotplate, leading to measurement errors from unexpected resistance changes. These errors can compromise temperature uniformity by disrupting the temperature control system. However, detecting sensor failures using data from a single sensor is challenging due to these measurement errors. This paper presents a real-time sensor diagnosis method that utilizes a two-dimensional image feature based on the sensor locations on the hotplate. The 2-D image feature is generated by mapping temperature data to the actual sensor locations on the hotplate layout, capturing thermal distribution information. A convolutional neural network (CNN) extracts the underlying thermal distribution. A failure injection test simulates various sensor failure scenarios based on known failure mechanisms. The CNN-based diagnosis method effectively classifies failure locations and the extent of the failures. Ultimately, the proposed method enables real-time diagnosis of RTDs embedded in a hotplate through a simple sensor location-based conversion.