Published Work

As a faculty member at Texas A&M University and Purdue University, I published in the areas of process modeling, spatial systems and artificial intelligence as applied to soil erosion and runoff, animal behavior, food process equipment controllers, robotic fruit harvesting and non-invasive measures of meat quality. All of this work was tied together by machine learning methods including pattern recognition and neural networks.

Principles of ultrasound and measurement of intramuscular fat

A review of basic engineering concepts of ultrasound is presented for the layperson with implications toward the use of ultrasound on beef animals. The use of ultrasound for determining quality traits such as percentage of intramuscular fat is discussed in detail. Results of both A-mode and B-mode preliminary investigations are presented. Preliminary results show that intramuscular fat may be predicted using an A-mode transducer coupled with frequency analysis. View Work >>

Ultrasonic Spectral Analysis for Beef Sensory Attributes

Ultrasonic spectral feature analysis was conducted for measuring beef sensory attributes noninvasively. Spectral features were compared with instrumental texture, chemical and sensory evaluation measures. The most significant (P <0.05) ultrasonic parameter was the number of local maxima for juiciness (ρ=0.49), connective tissue amount (ρ=0.52), flavor intensity (ρ=0.39), percent total collagen (ρ=0.34), and shear force (ρ=0.51). However, the central (resonant) frequency was the most dominant parameter for tenderness (ρ=0.45; P <0.05). Multivariate linear regression models were developed for predicting each palatability attribute. Standard errors of calibration for models were 0.253 for juiciness, 0.745 for muscle fiber tenderness, 0.244 for connective tissue amount, 0.754 for overall tenderness, and 0.224 for flavor intensity. Accuracy of prediction models was not adequate for use as a tool but this approach has potential for nondestructive sensory attribute measurement. View Work >>

Neural Network Prediction Modeling for a Continuous, Snack Food Frying Process

Automatic control is a primary concern of a continuous, snack food frying process. For the purpose of controlling product quality, two neural network paradigms were applied to develop prediction models to deal with the complexity of the process. Based on the modeling assumptions of the process, the neural network one-step-ahead and multiple-step-ahead predictors were established mathematically, the training algorithms for the two network predictors were developed, and a procedure for network prediction model identification was established. Results of model identification and predictions of the continuous, snack food frying process were presented in one-step-ahead and multiplestep-ahead modes. Prediction models developed in this article are ready for development of control loops. View Work >>