Sensor fusion & data analytics/machine learning for smart manufacturing

Applications in additive manufacturing. The research in this area is motivated by the existing challenges in quality assurance for additive manufacturing (AM). Although AM is bringing transformative impacts to industries, its sustainability is being constrained by the inherent limitation imposed by the layer-by-layer manner of fabrication that leads to numerous defects[J2]. To address this challenge, we developed a number of online in situ sensor-based data analytics methods to detect and then mitigate the onset of AM process defects[J13][J17][C2][C3][C5]. We also developed methods using offline measurement data for quality inspection of AM parts. For instance, we used point cloud data from 3D scanning to quantify the dimensional quality of AM parts with complex geometries[J8][J16]. We also devised a layer-wise spatial porosity model to quantify the porosity distribution based on CT scan data[J3].
Related literature on this topic is limited, I believe I have contributed significantly to this new area, in particular applications of advanced data analytics (such as machine learning) to AM quality control. As evidence of this contribution, I have received two NSF awards (projects #4,9 in Sec. V.C.a), which are the first of their kind, and three best paper awards[J13][C3][C5]. I have also been awarded by the prestigious ONR multidisciplinary research initiative (MURI) program for AM research (project #1 in Sec. V.C.a) as part of a six-university team.

Applications in semiconductor manufacturing. My research in this area is motivated by existing challenges in quality improvement for Chemical Mechanical Planarization (CMP), which is one of the most critical operations in wafer fabrication for semiconductor manufacturing. The challenges are two-fold[J19]: (1) complex CMP process dynamics; and (2) inherent uncertainties in sensor data.
I addressed the first challenge by investigating the relationship between the CMP performance, the process parameters, and the process state[J39], and by developing a method that integrates nonlinear Bayesian analysis and statistical modeling to predict performance measures[J28][J30]. To tackle the second challenge, I implemented an approach to effectively quantify sensor data uncertainty in CMP processes using a mixture Gaussian model[J9]. In addition, I devised a profile-based, statistical process control model for accurately monitoring wafer thickness profiles with non-normality in an industrial wafer slicing process[J4].
My research in this area has resulted in several effective methods to tackle CMP process complexity and uncertainty, which are expected to be beneficial to the semiconductor industry in efforts to improve wafer fabrication quality. Our research in this area has been recognized by one best paper award[J18], and three NSF awards (projects #12,14,15 in Sec. V.C.a).