Graduate Thesis Or Dissertation

Multi-modal sensing data fusion for machining and additive manufacturing

Public Deposited

Downloadable Content

Download PDF


Attribute NameValues
  • With the increased demand for high-quality products, reliable and robust process monitoring has become a key capability for modern manufacturing systems. Through sensors installed in the machinery, the real-time information was measured to assess the state of the processes. The measured data can be used to infer the mechanical, material, and geometric properties of the products, as well as the health status of the machining system. The direct and indirect measurements also provide real-time feedback for the control of the manufacturing process, thereby mitigating the risk of producing defective products or damaging machinery components. A multi-modal sensing system is one category of measurement technique developed in the last few decades. The system monitors the target through multiple physical perspectives to overcome the limitations of the single-modal sensing system. The captured physical parameters are then integrated into a representational form through a particular data processing or fusion algorithm. Due to its high accuracy, robustness, and reliability, the multi-modal sensing method has been widely applied in industries, including aviation, medical, and manufacturing systems. This thesis presents the theoretical and experimental study of multi-modal sensing approaches for the in-situ monitoring of two types of manufacturing systems: additive manufacturing and metal lathing. Specifically, the work included in this thesis is introduced in a total of four topics. The first three topics develop sensing techniques for monitoring the Laser Powder Bed Fusion (LPBF) process to estimate the material composition and defect formation during printing based on multi-physics thermal signatures extracted from Infrared images, while the last topic introduced a novel data fusion scheme for predicting the tool wear status base on multi-modal sensing. The first topic presented an in-situ material composition detection approach. This composition monitoring method was developed for poly-metal AM printing through thermal imaging by measuring the cooling rate of the melt pool during the solidification process. An analytical model is employed to address the correlation between the cooling rate, thermal properties of the material, and the processing parameters, which is subsequently utilized for retrieving the material composition. The analytical model has been implemented to simulate the printing process of copper-Inconel 625 alloys, and it predicts the monotonic relationship between the cooling rate and the copper’s weight percentage. To validate the relationship, a high-speed IR camera was adopted to monitor the printing process on a commercial Laser Powder Bed Fusion (L-PBF) system. A comparison between the experimental data and theoretical results revealed a firm agreement. The conclusive findings demonstrate the potential of the developed method to supervise the real-time alloy composition, thereby enhancing the consistency of mechanical and thermal properties in the printed parts. The second topic addressed a multi-modal sensing approach for real-time defect detection during L-PBF processes. The new method measures the cooling rate at a specific temperature and melt pool temperature of each location on the powder bed and then synthesizes the results into images representing the defect-induced thermal property variation on each printed layer. Analytical models have been built based on a simplified L-PBF process scenario to theoretically prove the nonlinear relationship between the thermal signatures and the voids-induced conductivity reduction. The relationship is then validated in experiments on a commercial L-PBF platform with a high-speed IR sensor. Ex-situ X-ray CT scanning was employed to validate the detection results of SynTG. The comparison shows a good match between the SynTG and X-ray CT Scanning results. The conclusive findings demonstrate the potential of the SynTG method in defect detection, thereby enhancing the quality of the printed parts. The third topic is an extension application of Topic 2, where a multi-observation Hidden Markov Model (HMM) was generated to predict defect formation during the laser powder bed fusion process by analyzing the thermal signatures. Being different from the second topic, this work considers the sequential printing process during LPBF, when the parts are built from bottom to top by following a layer-by-layer motion. This approach models the process as a Markov chain, where each layer print is considered a “state.” Based on the Markov chain model, “the future state is determined by the previous state.” Such a relationship was used to mathematically describe the physical fact that the quality of the lower layers could affect the upper layer properties at a certain statistical probability. Among the mathematical representations of Markov Chains, a Hidden Markov Model was selected and utilized to statistically analyze the time domain and space domain thermal information to estimate the invisible porosity of each layer. The thermal data and X-ray CT scanning results of Topic 2 were used to determine the parameters of the HMM and test the prediction capabilities. The HMM was separately trained with three sets of observation variables: 1) peak temperature, 2) cooling rate at a specific temperature level, and 3) multi-observation combining peak temperature and cooling rate at a specific temperature. The results show that, with multiple observation variables, the prediction accuracy reaches 92.7%, surpassing the accuracy of both single-observation HMMs. The improved accuracy indicates that multiple observation variables enhance the robustness, accuracy, and reliability of the HMM in predicting the formation of defects during the laser powder bed fusion processes. The last topic of this dissertation is the research and development of a multi-modal sensing system for predicting the tool wear state during the metal turning process. The study was based on a “smart tool” design, which integrated tool-temperature, vibration, strain (cutting force), and acoustic emission sensors in the turning tool shank to monitor the in-process parameters. This work developed a novel thermal signal-enhanced prediction method to infer the tool wear status based on Unscented Kalman Filter (UKF). Being different from the conventional UKF, the new prediction method introduced the tool temperature as a state variable correlated with the tool wear status and used it to update the state transition function in UKF for improved adaptability and accuracy. This method was validated with the smart tool through cutting tests on a commercial digital turning platform. The results show that the thermal signature-enhanced prediction algorithm improves the accuracy by 8% compared to the conventional UKF. 
Resource Type
Date Issued
Degree Level
Degree Name
Degree Field
Degree Grantor
Commencement Year
Academic Affiliation
Rights Statement
Peer Reviewed



This work has no parents.

In Collection: