Search NASA⌕ Search

SEARCH · Search NASA

Results for “Process Monitoring”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 775 records · Page 43

Algal Lipids and Omega-3 Production via Autotrophic and Heterotrophic Pathways at Cellana?s Kona Demonstration Facility, Hawaii

Cellana?s Kona Demonstration Facility (KDF) is a 2.5 hectare facility, with 17,000 sq. ft. under roof and 1 hectare of cultivation systems. KDF is designed to execute and support all stages of the production process at pilot scale, from cultivation through extraction. Since Feb. 2009, KDF has been producing up to 0.7MT dry weight of algal biomass per month, while at the same time optimizing processes of cultivation, harvesting, dewatering and extraction. The cultivation system at KDF uses ALDUO? technology, a hybrid system of photobioreactors (PBRs) and open ponds. All fluid transfers related to KDF cultivation and harvesting processes are operated and monitored by a remote Process-Control System. Fluid transfer data, together with biochemical data, enable the mass balance calculations necessary to measure productivity. This poster summarizes methods to improve both biomass and lipids yield by 1) alleviating light limitation in open ponds, 2) de-oxygenation and 3) heterotrophic lipid production for post-harvesting cultures.

09 BIOMASS FUELS↗

Exposure assessment of process-related contaminants in food by biomarker monitoring

Exposure assessment is a fundamental part of the risk assessment paradigm, but can often present a number of challenges and uncertainties. This is especially the case for process contaminants formed during the processing, e.g. heating of food, since they are in part highly reactive and/or volatile, thus making exposure assessment by analysing contents in food unreliable. New approaches are therefore required to accurately assess consumer exposure and thus better inform the risk assessment. Such novel approaches may include the use of biomarkers, physiologically based kinetic (PBK) modelling-facilitated reverse dosimetry, and/or duplicate diet studies. This review focuses on the state of the art with respect to the use of biomarkers of exposure for the process contaminants acrylamide, 3-MCPD esters, glycidyl esters, furan and acrolein. From the overview presented, it becomes clear that the field of assessing human exposure to process-related contaminants in food by biomarker monitoring is promising and strongly developing. The current state of the art as well as the existing data gaps and challenges for the future were defined. They include (1) using PBK modelling and duplicate diet studies to establish, preferably in humans, correlations between external exposure and biomarkers; (2) elucidation of the possible endogenous formation of the process-related contaminants and the resulting biomarker levels; (3) the influence of inter-individual variations and how to include that in the biomarker-based exposure predictions; (4) the correction for confounding factors; (5) the value of the different biomarkers in relation to exposure scenario’s and risk assessment, and (6) the possibilities of novel methodologies. Here, in spite of these challenges it can be concluded that biomarker-based exposure assessment provides a unique opportunity to more accurately assess consumer exposure to process-related contaminants in food and thus to better inform risk assessment.

63 RADIATION, THERMAL, AND OTHER ENVIRON. POLLUTAN↗

Spectroscopic Monitoring of Spent Nuclear Fuel Reprocessing Streams: An Evaluation of Spent Fuel Solutions via Raman, Visible, and Near-Infrared Spectroscopy

The potential of using optical spectroscopic techniques, such as Raman and Visible/Near Infrared (Vis/NIR), for on-line process control and special nuclear materials accountability applications at a spent nuclear fuel reprocessing facility was evaluated. Availability of on-line real-time techniques that directly measure process concentrations of nuclear materials will enhance performance and proliferation resistance of the solvent extraction processes. Further, on-line monitoring of radiochemical streams will also improve reprocessing plant operation and safety. This report reviews current state of development of the spectroscopic on-line monitoring techniques for such solutions. To further examine applicability of optical spectroscopy for monitoring reprocessing solutions, segments of a spent nuclear fuel, with approximate burn-up values of 70 MWd/kgM, were dissolved in concentrated nitric acid and adjusted to varying final concentrations of HNO3. The resulting spent fuel solutions were batch-contacted with tributyl phosphate/dodecane organic solvent. The feed and equilibrium aqueous and loaded organic solutions were subjected to optical measurements. The obtained spectra showed the presence of the quantifiable Raman bands due to NO3- and UO22+ and Vis/NIR bands due to multiple species of Pu(IV), Pu(VI), Np(V), the Np(V)-U(VI) cation-cation complex, and Nd(III) in fuel solutions, justifying spectroscopic techniques as a promising methodology for monitoring spent fuel processing solutions in real-time. Quantitative evaluation of the fuel solution was performed based on spectroscopic measurements and compared to ICP-MS analysis.

11 NUCLEAR FUEL CYCLE AND FUEL MATERIALS↗

Near infrared laser annealing of CdTe and in-situ measurement of the evolution of structural and optical properties

The high performance of polycrystalline CdTe thin film solar cells is enabled by annealing in the presence of Cl. This process is typically carried out for tens of minutes resulting in reduction of defect states within the bandgap among other beneficial effects. In this work, we investigate laser annealing as a means of rapidly annealing CdTe using a continuous wave sub-bandgap 1064 nm laser. The partial transmission of the beam allows us to monitor the annealing process in-situ and in real time. We find that optoelectronic and structural changes occur through two distinct kinetic processes resulting in the removal of deep defects and twinned regions, respectively. A multilayer optical model including surface roughness is used to interpret both the in-situ transmission as well as ex-situ reflectivity measurements. These experiments demonstrate beneficial material changes resulting from sub-bandgap laser-driven CdCl{sub 2} treatment of CdTe in minutes, which is an important step towards accelerating the processing of the CdTe absorber layer.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Plasma nitriding monitoring reactor: A model reactor for studying plasma nitriding processes using an active screen

A laboratory scale plasma nitriding monitoring reactor (PLANIMOR) has been designed to study the basics of active screen plasma nitriding (ASPN) processes. PLANIMOR consists of a tube reactor vessel, made of borosilicate glass, enabling optical emission spectroscopy (OES) and infrared absorption spectroscopy. The linear setup of the electrode system of the reactor has the advantages to apply the diagnostic approaches on each part of the plasma process, separately. Furthermore, possible changes of the electrical field and of the heat generation, as they could appear in down-scaled cylindrical ASPN reactors, are avoided. PLANIMOR has been used for the nitriding of steel samples, achieving similar results as in an industrial scale ASPN reactor. A compact spectrometer using an external cavity quantum cascade laser combined with an optical multi-pass cell has been applied for the detection of molecular reaction products. This allowed the determination of the concentrations of four stable molecular species (CH{sub 4}, C{sub 2}H{sub 2}, HCN, and NH{sub 3}). With the help of OES, the rotational temperature of the screen plasma could be determined.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗

IACMI Project 4.2: Thermoplastic Composite Development for Wind Turbine Blades

(Section 5.1) Composites made from Arkema’s Elium® thermoplastic resin and Johns Manville fiberglass were researched during this project for applications in wind blade manufacturing. A techno-economic model was developed to model this wind blade manufacturing process using these materials in place of traditional composites made with thermoset resin. This model was based on manufacturing a 61.5-meter wind blade, which showed a 4.7% reduction in wind blade cost as compared traditional thermoset materials. These cost savings were not from the thermoplastic material costing less than traditional thermoset materials, but rather from decreased capital costs, faster cycle times and reduced energy requirements and labor costs. (Section 5.2) An infusion and curing model was developed for thermoplastic composite wind blades using PAM-RTM. The primary goal was to demonstrate the infusion simulation for the Elium® resin system on a 13-meter wind blade. Additionally, the exotherm temperature was predicted and compared to measurements, which showed model results within 10% of actual measurements. (Section 5.3) Composite laminate panels and composite sandwich panels with a balsa core were produced; specimens were cut and characterized. Similar composite specimens were made with Elium® thermoplastic resin and Hexion thermoset epoxy (RIMR135/RIMH1366) to enable comparisons between these resin systems. The static test methods included: tensile, compression, in-plane shear, interlaminar shear, flexural, sandwich core shear flexure, and single cantilever beam tests for sandwich beams. Fatigue testing at room temperature was completed to composite laminate panels at a stress ratio of R=0.1 and R=10. In addition, fatigue testing to laminate panels was completed at -30°C, and at room temperature after conditioning specimens at 70°C and 90% relative humidity. Overall, mechanical test results from Elium® composites are similar to epoxy composites. (Section 5.4) Elium composite panels were produced with intentional defects such as voids and nonwetting of fibers to begin to understand performance sensitivity to defects. A thermal digital image correlation (TDIC) method provides high spatial resolution strain field at elevated temperatures and can be used to identify defective regions within composite panels. Flexural modulus differences of 21% were seen between defect and non-defect panels. Other Elium® composite panels were forced to be defective by boiling the resin after infusion, which created voids throughout the composite laminate. X-ray computed tomography scanning was used to view the internal structure of the defect panels. Defect panels had a significant reduction in fatigue life as compared to baseline panels produced without intentional defects. (Section 5.5) Lap shear specimens were fabricated to compare the lap shear strength of an off-the-shelf adhesive (Plexus MA590) and two new adhesives developed by Arkema (Bostik SAF30 90 and Bostik SAF30 120). ISO standard 4587:2003 was used to standardize the testing method and sample fabrication. Lap shear specimens were made at 1mm, 3mm, and 10mm thicknesses. The Bostik adhesive lap shear test results were similar to Plexus for all thicknesses. (Section 5.6) Fiber-reinforced polymer (FRP) composites are typically used in high-performance applications (e.g., aerospace), and their expansion into high-volume industries (e.g. consumer automotive and wind turbine blade manufacturer or similar) is hindered by their cost and a lack of efficient manufacturing techniques. Monitoring the curing process of these composites during manufacturing can improve the efficiency of the process, and therefore reduce the manufacturing cost. Cure monitoring techniques were developed that use probabilistic estimation methods and surface temperature measurements made using infrared cameras. These techniques enable real-time monitoring of the infusion process to locate manufacturing flaws, and they can, potentially, estimate residual stresses in the part. Their commercialization will help facilitate expansion of FRP composites in high-volume industries. (Section 5.7) A 13-meter composite wind blade was produced with Elium® resin and Johns Manville fiberglass; this blade was made with VARTM processing similar to how megawatt-scale wind blades are currently manufactured, but no post-mold heating was used for this thermoplastic composite blade. The wind blade underwent full-scale validation for static loading (4-different load orientations) and flapwise fatigue loading to simulate 20-years of operational loads. The thermoplastic composite wind blade withstood the loading without any noted issues and performed similar to results from a previous full-scale validation to an equivalent epoxy composite wind blade produced with the same blade molds. (Section 5.8) A study was conducted to determine the feasibility of recycling composite wind turbine blade components fabricated with glass fiber reinforced Elium® thermoplastic resin. Dissolution, which is a process unique to thermoplastic matrices, allows recovery of both the polymer matrix and full-length glass fibers, while maintaining their stiffness and strength throughout the recovery process. The economics of recycling is favorable if 50% of the glass fiber is recovered and resold for a process of $\$$ 0.28/kg, and 90% of the resin is recovered and resold at a price of $\$$ 2.50/kg.(Section 10) Recommendations are outlined for commercializing thermoplastic resin for composite wind blade production, in addition to recommended areas for future research.

17 WIND ENERGY↗

Real-time laser ultrasonic monitoring of laser-induced thermal processes

Abstract Intra- and inter-layer integrity of components fabricated with advanced manufacturing techniques, such as laser powder bed fusion, is dependent upon rapid heating, melting, and solidification processes. There is a need for new techniques to provide in situ feedback of these processes. Here a laser-based ultrasonic technique to probe thermal effects induced by a high-power continuous wave laser in titanium samples is described. Numerical simulations were performed to show that, for a spatially uniform heating beam, laser-induced surface acoustic waves are strongly influenced by surface heating conditions, are dispersive in the case of rapid heating, and that an abrupt velocity reduction happens upon the onset of surface melting. Furthermore, laser-based ultrasound experimental results which monitor the transient change of surface wave travel time associated with high power laser surface heating are provided. A pulsed laser is used to generate high frequency surface acoustic waves that propagate through the laser-heated region and are detected using a photorefractive crystal-based interferometer. Qualitative agreement is observed between theory and experiment with both showing a rapid reduction in the surface wave velocity at the onset of illumination and further decrease in surface wave velocity associated with melting. It is demonstrated that changes in the surface wave velocity can be used to track local heating and detect the onset of surface melting in real time.

36 MATERIALS SCIENCE↗

An experimental set-up to apply polarization modulation to infrared reflection absorption spectroscopy for improved in situ studies of atmospheric corrosion processes

A new set-up for improved monitoring of atmospheric corrosion processes in situ and in real-time is presented. To characterize chemical structures of thin films on metal surfaces surface sensitive analytical techniques are required. One possible technique is Infrared Reflection Absorption Spectroscopy (IRRAS) which has become an established method to investigate surface corrosion films of thicknesses less than 200 nm. However, there are limitations related to the sensitivity of these measurements, in case of investigating ultrathin films or absorption bands of interest, surface species are superimposed by atmospheric background absorption, which changes during in situ measurements in ambient atmospheres. These difficulties of in situ surface reflection measurements can be eliminated by availing the polarization selectivity of adsorbed surface species. At grazing angles of incidence the absorption of p-polarized infrared radiation by thin surface films on metals is enhanced, while the absorption of s-polarized light by this film is nearly zero. This different behavior of the polarization properties leads to strong selection rules at the surface and can therefore be used to identify molecules adsorbed on metal surfaces. Polarization Modulation (PM) of the infrared (IR) light takes advantage of this disparity of polarization on sample surfaces and in combination with IRRAS yielding a very sensitive and surface-selective method for obtaining IR spectra of ultra-thin films on metal surfaces. An already existing in situ IRRAS/Quartz Crystal Microbalance weathering cell was combined with PM and evaluated according to its applicability to study in situ atmospheric corrosion processes. First real-time measurements on silver samples exposed to different atmospheres were performed showing the advantage of PM-IRRAS compared to conventional IRRAS for such investigations.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗

Three Years of Multilevel Monitoring Data at the Riverton, Wyoming, Processing Site That Show Contaminant Increases After River Flooding Events and a Large Recharge Event

The tailings pile at the Riverton, Wyoming, Processing Site has been removed, but contaminant concentrations in the shallow aquifer increase after large recharge events. The Riverton site experienced flooding near the Little Wind River in early May 2016, February 2017, and June 2017. The May 2016 flooding was from a large rain event. The February 2017 flooding was due to an ice jam, and the June 2017 flooding was due to mountain snowmelt runoff. In addition, the site experienced high water tables in April and May 2017, as this was the second wettest spring in Riverton since records began in 1918. Past reports indicated the accumulation of solid-phase uranium and molybdenum in evaporites above the respective contaminant groundwater plumes. Given these data, multilevel monitoring wells were installed in 2015 to monitor seasonal changes in groundwater quality. Data from three years of groundwater monitoring are summarized in this report and compared with the flooding and large recharge events. The results indicate that evaporite-related constituents (especially sodium, sulfate, and chloride) present across the site in a silt layer can be released to the shallow sand and gravel aquifer during flooding and large recharge events. The solid-phase and water-phase data confirm that these evaporites are mainly sodium sulfate and sodium chloride salts with some calcite and gypsum.

12 MANAGEMENT OF RADIOACTIVE AND NON-RADIOACTIVE W↗

$\mathrm{RADICAL}$-Pilot and $\mathrm{PMIx}$/$\mathrm{PRRTE}$: Executing Heterogeneous Workloads at Large Scale on Partitioned $\mathrm{HPC}$ Resources

Execution of heterogeneous workflows on high-performance computing (HPC) platforms present unprecedented resource management and execution coordination challenges for runtime systems. Task heterogeneity increases the complexity of resource and execution management, limiting the scalability and efficiency of workflow execution. Re-source partitioning and distribution of tasks execution over portioned re-sources promises to address those problems but we lack an experimental evaluation of its performance at scale. Here this paper provides a performance evaluation of the Process Management Interface for Exascale (PMIx) and its reference implementation PRRTE on the leadership-class HPC plat-form Summit, when integrated into a pilot-based runtime system called RADICAL-Pilot. We partition resources across multiple PRRTE Distributed Virtual Machine (DVM) environments, responsible for launching tasks via the PMIx interface. We experimentally measure the work-load execution performance in terms of task scheduling/launching rate and distribution of DVM task placement times, DVM startup and termination overheads on the Summit leadership-class HPC platform. Integrated solution with PMIx/PRRTE enables using an abstracted, standardized set of interfaces for orchestrating the launch process, dynamic process management and monitoring capabilities. It extends scaling capabilities allowing to overcome a limitation of other launching mechanisms (e.g., JSM/LSF). Explored different DVM setup configurations provide insights on DVM performance and a layout to leverage it. Our experimental results show that heterogeneous workload of 65,500 tasks on 2048 nodes, and partitioned across 32 DVMs, runs steady with resource utilization not lower than 52%. While having less concurrently executed tasks resource utilization is able to reach up to 85%, based on results of heterogeneous workload of 8200 tasks on 256 nodes and 2 DVMs.

97 MATHEMATICS AND COMPUTING↗

Atomic-scale electrochemistry on the surface of a manganite

The doped manganese oxides (manganites) have been widely studied for their colossal magnetoresistive effects, for potential applications in oxide spintronics, electroforming in resistive switching devices, and are materials of choice as cathodes in modern solid oxide fuel cells. However, little experimental knowledge of the dynamics of the surfaces of perovskite manganites at the atomic scale exists. Here, through in-situ scanning tunnelling microscopy (STM), we demonstrate atomic resolution on samples of La 0.625 Ca 0.375 MnO 3 grown on (001) SrTiO 3 by pulsed laser deposition (PLD). Furthermore, by applying triangular DC waveforms of increasing amplitude to the STM tip, and measuring the tunnelling current, we demonstrate the ability to both perform and monitor surface electrochemical processes at the atomic level, including, for the first time in a manganite, formation of single and multiple oxygen vacancies, disruption of the overlying manganite layers, and removal and deposition of individual atomic units or clusters. Our work paves the way for better understanding of surface oxygen reactions in these systems.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Reversed Nanoscale Kirkendall Effect in Au–InAs Hybrid Nanoparticles

Metal–semiconductor hybrid nanoparticles (NPs) offer interesting synergistic properties, leading to unique behaviors that have already been exploited in photocatalysis, electrical, and optoelectronic applications. A fundamental aspect in the synthesis of metal–semiconductor hybrid NPs is the possible diffusion of the metal species through the semiconductor lattice. The importance of understanding and controlling the co-diffusion of different constituents is demonstrated in the synthesis of various hollow-structured NPs via the Kirkendall effect. Here, we used a postsynthesis room-temperature reaction between AuCl 3 and InAs nanocrystals (NCs) to form metal–semiconductor core–shell hybrid NPs through the “reversed Kirkendall effect”. In the presented system, the diffusion rate of the inward diffusing species (Au) is faster than that of the outward diffusing species (InAs), which results in the formation of a crystalline metallic Au core surrounded by an amorphous, oxidized InAs shell containing nanoscale voids. We used time-resolved X-ray absorption fine-structure (XAFS) spectroscopy to monitor the diffusion process and found that both the size of the Au core and the extent of the disorder of the InAs shell depend strongly on the Au-to-NC ratio. We have determined, based on multielement fit analysis, that Au diffuses into the NC via the kick-out mechanism, substituting for In host atoms; this compromises the structural stability of the lattice and triggers the formation of In–O bonds. These bonds were used as markers to follow the diffusion process and indicate the extent of degradation of the NC lattice. Time-resolved X-ray diffraction (XRD) was used to measure the changes in the crystal structures of InAs and the nanoscale Au phases. By combining the results of XAFS, XRD, and electron microscopy, we correlated the changes in the local structure around Au, As, and In atoms and the changes in the overall InAs crystal structure. This correlative analysis revealed a co-dependence of different structural consequences when introducing Au into the InAs NCs. Therefore, this study of diffusion effects in nanocrystals has relevance to powerful concepts in solid-state nanochemistry related to processes of cation exchange, doping reactions, and diffusion mechanisms.

36 MATERIALS SCIENCE↗

Raman Spectroscopy Based On-Line, Real-Time Monitoring to Reduce Composition Uncertainties: Enhanced sensitivity through optimization of Raman Parameters

Optical spectroscopy-based on-line monitoring of Hanford processing streams can enable real-time characterization of chemical composition of process streams and batches, ultimately enabling and enhancing process control. It can provide immediate feedback on process conditions and has the potential to reduce the needed number of grab sample collections, thereby reducing times and costs associated with laboratory processing. Here we discuss the utilization of Raman spectroscopy to quantify multiple target analytes that are common within Hanford tanks and waste processing streams. Analytes include: nitrate, nitrite, carbonate, chromate, sulfate, phosphate, hydroxide, oxalate, ammonia, and aluminate. Most notably in this work, Raman applications to low-concentration streams are explored and optimized. Raman instrument specifications are compared; specifically, the impact of utilizing three different Raman excitation wavelengths, 405, 532, and 671 nm, is discussed. Also, Raman data collection parameters such as collection time and spectral averaging are measured and discussed. Finally, optical libraries of chemical targets were collected using optimized collection parameters and chemometric models were built to automate quantification of chemical targets. These models were validated through application to simulants and real Hanford process samples. Chemometric models performed well on both training and validation sets, suggesting these approaches can be successfully applied to on-line and real-time monitoring of low concentration Hanford processing streams. The use of the combine Raman wave lengths with the enhanced chemometric models for the low concentration streams significantly improved the chemical detection levels and significantly reduced uncertainties of those measurements.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗

Explainable AI (XAI)-driven vibration sensing scheme for surface quality monitoring in a smart surface grinding process

Local Interpretable and Model-agnostic Explanation (LIME), an explainable artificial intelligence (XAI) approach is adapted to identify the globally important time-frequency bands for predicting average surface roughness (Ra) in a smart grinding process. The smart grinding setup consisted of a Supertech CNC precision surface grinding machine, instrumented with a Dytran piezoelectric accelerometer attached to the tailstock quill along the tangential direction (Y-axis). For every grinding pass, vibration signatures were captured, and the ground truth surface roughness values were recorded using a Mahr Marsurf M300C portable surface roughness profilometer. The roughness values ranged from 0.06 to 0.14 microns over the complete set of experiments. Time-frequency domain spectrogram frames were extracted for each of the vibration signals collected during the grinding process. Convolutional Neural Networks (CNNs) were modeled to predict the surface roughness based on these spectrogram frames and their image augmentations. The best CNN model was able to predict the roughness values with an overall R2-score of 0.95, training R2-score of 0.99, and testing R2-score of 0.81 with only 80 sets of vibration signals corresponding to 4 experiments with 20 trials each. Although the data size is not large enough to guarantee such performance metrics in real-world scenarios, one can extract statistically consistent explanations underlying the relationships these complex deep learning models capture. Further, the LIME methodology was implemented on the developed surface roughness CNN model to identify the important time-frequency bands (i.e., the superpixels of a spectrogram) influencing the predictions. Based on the identified important regions on the spectrogram frames, the corresponding frequency characteristics were determined that influence the surface roughness predictions. The important frequency range based on LIME results was approximately 11.7 to 19.1 kHz. The power of XAI was demonstrated by cutting down the sampling rate from 160 kHz to 30, 20, 10, and 5 kHz based on the important frequency range and considering Nyquist criteria. Separate CNN models were developed for these ranges by only extracting time-frequency contents below their corresponding Nyquist cut-offs. A proper data acquisition strategy is proposed by comparing the model performances to argue the selection of a sufficient sampling rate to capture the grinding process successfully and robustly.

42 ENGINEERING↗

Reprint of: Explainable AI (XAI)-driven vibration sensing scheme for surface quality monitoring in a smart surface grinding process

Local Interpretable and Model-agnostic Explanation (LIME), an explainable artificial intelligence (XAI) approach is adapted to identify the globally important time-frequency bands for predicting average surface roughness (Ra) in a smart grinding process. The smart grinding setup consisted of a Supertech CNC precision surface grinding machine, instrumented with a Dytran piezoelectric accelerometer attached to the tailstock quill along the tangential direction (Y-axis). For every grinding pass, vibration signatures were captured, and the ground truth surface roughness values were recorded using a Mahr Marsurf M300C portable surface roughness profilometer. The roughness values ranged from 0.06 to 0.14 microns over the complete set of experiments. Time-frequency domain spectrogram frames were extracted for each of the vibration signals collected during the grinding process. Convolutional Neural Networks (CNNs) were modeled to predict the surface roughness based on these spectrogram frames and their image augmentations. The best CNN model was able to predict the roughness values with an overall R2-score of 0.95, training R2-score of 0.99, and testing R2-score of 0.81 with only 80 sets of vibration signals corresponding to 4 experiments with 20 trials each. Although the data size is not large enough to guarantee such performance metrics in real-world scenarios, one can extract statistically consistent explanations underlying the relationships these complex deep learning models capture. Further, the LIME methodology was implemented on the developed surface roughness CNN model to identify the important time-frequency bands (i.e., the superpixels of a spectrogram) influencing the predictions. Based on the identified important regions on the spectrogram frames, the corresponding frequency characteristics were determined that influence the surface roughness predictions. The important frequency range based on LIME results was approximately 11.7 to 19.1 kHz. The power of XAI was demonstrated by cutting down the sampling rate from 160 kHz to 30, 20, 10, and 5 kHz based on the important frequency range and considering Nyquist criteria. Separate CNN models were developed for these ranges by only extracting time-frequency contents below their corresponding Nyquist cut-offs. A proper data acquisition strategy is proposed by comparing the model performances to argue the selection of a sufficient sampling rate to capture the grinding process successfully and robustly.

47 OTHER INSTRUMENTATION↗