Current Projects

Physics-Guided Data Learning Models (Greis, Nogueira, Bhattacharya, Cherukuri)

The research objective is to combine data-driven and physics-based, hybrid, models with process measurements to realize physics-guided data learning models. To improve the predictions, physics-based assessments can also be included to penalize results that are inconsistent with physical knowledge. By assuring physical consistency of the model predictions, the capability to generalize to other situations is increased. This enables the physically meaningful model output to be incorporated in new scientific discovery efforts, such as improvements in physics-based models.

Hybrid models are a relatively unexplored multi-disciplinary methodology. Successful integration of both models represents a transformative breakthrough in intelligent system development for manufacturing, broadly, and self-aware machining and metrology, in particular.

Chatter Avoidance (Cherukuri, Greis, Nogueira, Bhattacharya, Selles, Perez-Bernabeu)

In machining, chatter is a self-excited vibration that affects the surface finish and reduces material removal rate, spindle life, and tool life. The Stability Lobe Diagrams (SLDs) where the critical chip width that separates chatter from no chatter is plotted as a function of the spindle speed are often used to determine the machining parameters for maximal material removal rates without chatter. However, chatter prediction using SLDs has some well-known limitations. For this reason, several researchers have recently started working on the application of machine learning methods to predict chatter for a given set of machining parameters.

In this project, the objective is to blend artificial intelligence (AI), physics-based machining models, and in-process data to achieve self-aware machining performance. The approach involves using the SLD curves and machine learning to build the initial predictive models. Experiments will then be used to inform and update these physics-guided models. When implemented in a real machining process on the shop floor, the models can potentially aid in the online adjustment of the machining parameters to avoid chatter while at the same time, undergoing self-updates.

Data-Driven Metrology ( Falaggis, Bhattacharya)

Modeling complex systems that describe cross-disciplinary phenomena is difficult, and in some cases, impossible if governing physical equations are unknown. Data-driven methods discover underlying principles from data without any knowledge about the governing equations. This project employs machine-learning techniques to obtain self-generated, data-driven models for optical metrology in manufacturing that improve the accuracy of the measurement of capabilities for metrology approaches such as Fringe projection, camera calibration, holography and interferometry.

Ultra-Precision Machining of Freeform Optics (Davies, Suleski, Greis, Nogueira)

The research objective is to utilize physics-based models of machining and optical performance in concert with in-process monitoring to train learning systems and achieve self-aware ultra-precision machining of optics. Such a system would produce a competitive advantage by minimizing or eliminating current costly set-up and compensation procedures. The project tasks are:

Semantic Data Management for Manufacturing and Metrology (Morse, Greis)

Manufacturing relies on a variety of sensors to provide the data needed to maintain process control and product quality. The measurement data may be related to attributes of the product (dimensions, material properties, surface integrity) or the manufacturing equipment (temperature, vibration). Constructing a useful application of these data requires that the data are traceable – that is, the source of information, time of its collection, and other uniquely identifiable attributes are known and maintained. An application domain for this research will be automated point cloud data analysis. Classical coordinate metrology uses high precision structures (coordinate measuring machines, or CMMs) to collect point data with a spherical probe from a workpiece surface resulting in accurate, but sparse points; while light-based sensing gather hundreds of thousands of points per second resulting in nontrivial feature identification and extraction. The research objective is to provide a traceability framework for data-driven models and apply this framework to the analysis of the collected data from either spherical- or light-based probes.

Machining Force Diagnosis (Tarbutton)

The research objective is to enable self-awareness by combining information from a voxel-based machining force model with measured process signals. The hypothesis is that forces experienced during machining can be correlated to expected forces inferred through measured vibration, spindle current, and sound emission.

After a part is designed and ready to be machined, a process engineer uses computer aided manufacturing (CAM) software to generate the tool path. The process engineer is responsible for selecting the cutter geometry, cutting speed, feed rate, and depths of cut. This parameter selection is performed using handbooks, experience, cutting tool manufacturer recommendations, and software tools that attempt to aggregate this information into a simplified user interface. As such, parameters are often selected regardless of the part geometry and corresponding tool path. In practice, these recommended parameters encourage users to machine using sub-optimal conditions which results in hidden costs. A commercial response to this problem is the purchase of additional software that simulates the tool path and force levels. This simulation is completed offline without process feedback and does not permit self-aware behavior. The proposal of a self-aware machine enables the ability to compare the machining actual state during cutting to its expected state and makes intelligent decisions based on the difference.