Experimental Study Of Improved Transient Plane Source

Thermal conductivity is one of the important thermal properties of the material, widely used in building energy efficiency, petrochemical, aerospace, etc., fast and accurate measurement of the thermal conductivity of the material is significant. Transient Plane Source method is one of the thermal conductivity measurement method, with a wide measuring range, shorttime, easy to prepare specimens, etc; But, as ambient temperature and power fluctuation, and the contact thermal resistance affection, the measurement accuracy is low. In this paper, based on the transient plane source, established a BP neural network model of solid thermal conductivity, and an alternative to traditional BP algorithm with L-M algorithm to further improve measurement accuracy. Mainly to complete the following tasks:Theoretical analysis the transient thermal process of semi-infinite objects, and obtained temperature rise expression for sensor surface. Outlined the measuring principle of the modified TPS method, and deduced thermal conductivity formula, and analyzes the main sources of measurement error. Introduction the measuring principle of RTD, sensor’s material selection and size design, manufacture process specimens, etc., design of the sensor radius r = 6.4mm, thickness h = 60μm, the minimum diameter and minimum thickness of specimen is 4.4r and 1.2r. Introduction the basic principle of stable DC power supply, and in accordance with design requirements, selected Agilent E3642 A DC power supply for the auxiliary heating power. Standardized the designed sensor and calibrated the sampling resistor. Measuring the thermal conductivity of phosphor bronze, aluminum, Pyrex glass, LAF 6720 foam, ceramic, etc. by C-Therm TCi Thermal Conductivity. Considered heating power, heating time, density, ambient temperature, voltage change(or voltage change rate) and temperature rise(or temperature rise rate) as the key input variables, and established the BP network model of solid thermal conductivity, and using MATLAB7.5 to create, train, test and other processes for the network.Experimental Results: Model 1, Model 2 and Model 3’s respectively relative error is not exceed 3.5%, 2%, 10%, indicating that the network structure of model 2 is better than model 1 and model 3, and it’s worthy to recommend model 2 as a reference to establish BP network of solid thermal conductivity. It also shows that, the experimental program of using the BP network to reduce relative error for TPS method is feasible; Significant input variables affect the performance of the network.