GB1351091A - Thermal analysis method and a-paratus - Google Patents

Abstract

1351091 Thermal analysis GENERAL MOTORS CORP 18 May 1972 [27 May 1971] 23352/72 Heading G4A The metallurgical properties of a sample of molten cast-iron treated to form nodular castiron on solidification are predicted by (a) measuring the temperature of the sample at predetermined increments of time during an interval in which the sample cools from a liquid to a solid to determine a cooling curve, (b) storing the measurements in an electronic digital computer, (c) differentially analysing the measurements to determine (1) a first set of signals corresponding to the points of inflection of the cooling curve and (2) a second set of signals corresponding to points of minimum radius between the points of inflection and (d) storing the two sets of signals in matrix form as a function of independent variables such as absolute time and temperature. Known matrices obtained in similar manner from samples having known properties are correlated using regression analysis to determine the independent variables required to predict particular properties, these variables then being operated on to make predictions for the unknown sample. A sampling device (see Division H1) for determining the cooling curve includes a thermocouple assembly the output leads from which are coupled to a digital computer which stores the measured temperatures. The inflection points represent the temperatures at which the heat of solidification given by the samples are either a maximum or a minimum. They are computed by examining the first five measurements to determine the slope of the curve at the third measurement point, this process then being repeated for the second to sixth samples, &c. The maximum and minimum slopes are then selected to identify the points of inflection to derive the first set of signals. The points of minimum curve radius are computed by finding the minima of the equation where T is temperature; t is time to derive the second set of signals. These signals are stored in matrix form as functions of absolute time, temperature, temperature decay rate, curve radius and time and temperature differentials. Known matrices obtained in like manner from similar samples with known properties are correlated by a series of regressive analyses of the form where Y is the property to be predicted, A 0 is a constant dependent on this property, A 1 -A j are numerical coefficients and X 1 -X j are the stored independent variables selected by operating successively with each independent variable as X 1 to determine which gives the best prediction of the known property and then trying the remainder of the variables as X 2 &c. Predictions for the unknown sample are then effected by using the stored variables as the appropriate independent variables X i .