Dual-frequency comb spectroscopy has emerged as a disruptive technique for measuring wide-spanning spectra with high resolution, yielding a particularly powerful technique for sensitive ...multi-component gas analysis. We present a spectrometer based on two electro-optical combs with subsequent conversion to the mid-infrared via tunable difference frequency generation, operating in the range from 3 to 4.7 µm. The repetition rate of the combs can be tuned from 250 to 500 MHz. For 500 MHz, the number of detected comb modes is 440 with a signal-to-noise ratio exceeding 10 5 in 1 s. The conversion preserves the coherence of the combs within 3 s measurement time. Concentration measurements of 5 ppm methane at 3.3 µm, 100 ppm nitrous oxide at 3.9 µm and a mixture of 15 ppm carbon monoxide and 5% carbon dioxide at 4.5 µm are demonstrated with a noise-equivalent absorption coefficient of 6.4(3) x 10 −6 cm −1 Hz −1/2 .
Artificial neural networks (ANNs) are used in quantitative infrared gas spectroscopy to predict concentrations on multi-component absorption spectra. Training of ANNs requires vast amounts of ...labelled training data which may be elaborate and time consuming to obtain. Additional data can be gained by the utilization of synthetically generated spectra, but at the cost of systematic deviations to measured data. Here, we present two approaches to train ANNs with a combination of comparatively small, measured data sets and synthetically generated data. For the first approach a neural network is trained hybridly with synthetically generated infrared absorption spectra of mixtures of N
O and CO and measured zero-gas spectra, taken with a mid-infrared dual comb spectrometer. This improves the mean absolute error (MAE) of the network predictions from 0.46 to 0.01 ppmV and 0.24 to 0.01 ppmV for the concentration predictions of N
O and CO respectively for zero-gas measurements which was previously observed for training with purely synthetic data. At the same time a similar performance on spectra from gas mixtures of 0–100 ppmV N
O and 0 to 60 ppmV CO was achieved. For the second approach an ANN pre-trained on synthetic infrared spectra of mixtures of acetone and ethanol is retrained on a small dataset consisting of 26 spectra taken with a mid-infrared photoacoustic spectrometer. In this case the MAE for the concentration predictions of ethanol and acetone are improved by 45 % and 20 % in comparison to purely synthetic training. This shows the capability of using synthetically generated data to train ANNs in combination with small amounts of measured data to further improve neural networks for gas sensing and the transferability between different sensing approaches.
Abstract
A dual comb spectrometer is used as gas sensor for the parallel detection of nitrous oxide (N
2
O) and carbon monoxide (CO). These gases have overlapping absorption features in the ...mid-infrared (MIR) at a wavelength of 4.6 µm. With a spectra acquisition rate of 10 Hz, concentrations of 50 ppm N
2
O and 30 ppm CO are monitored with a relative precision of
6
×
10
−
3
6\times {10^{-3}}
and
3
×
10
−
3
3\times {10^{-3}}
respectively. The limit of detections are 91 ppb for N
2
O and 50 ppb for CO for an integration time of 25 s. The system exhibits a linear sensitivity from 2 ppm to 100 ppm with coefficients of determination of 0.99998 for N
2
O and 0.99996 for CO.
A dual comb spectrometer is used as gas sensor for the parallel detection of nitrous oxide (N
O) and carbon monoxide (CO). These gases have overlapping absorption features in the mid-infrared (MIR) ...at a wavelength of 4.6 µm. With a spectra acquisition rate of 10 Hz, concentrations of 50 ppm N
O and 30 ppm CO are monitored with a relative precision of
and
respectively. The limit of detections are 91 ppb for N
O and 50 ppb for CO for an integration time of 25 s. The system exhibits a linear sensitivity from 2 ppm to 100 ppm with coefficients of determination of 0.99998 for N
O and 0.99996 for CO.
Abstract
Artificial neural networks (ANNs) are used in quantitative infrared gas spectroscopy to predict concentrations on multi-component absorption spectra. Training of ANNs requires vast amounts ...of labelled training data which may be elaborate and time consuming to obtain. Additional data can be gained by the utilization of synthetically generated spectra, but at the cost of systematic deviations to measured data. Here, we present two approaches to train ANNs with a combination of comparatively small, measured data sets and synthetically generated data. For the first approach a neural network is trained hybridly with synthetically generated infrared absorption spectra of mixtures of N
2
O and CO and measured zero-gas spectra, taken with a mid-infrared dual comb spectrometer. This improves the mean absolute error (MAE) of the network predictions from 0.46 to 0.01 ppmV and 0.24 to 0.01 ppmV for the concentration predictions of N
2
O and CO respectively for zero-gas measurements which was previously observed for training with purely synthetic data. At the same time a similar performance on spectra from gas mixtures of 0–100 ppmV N
2
O and 0 to 60 ppmV CO was achieved. For the second approach an ANN pre-trained on synthetic infrared spectra of mixtures of acetone and ethanol is retrained on a small dataset consisting of 26 spectra taken with a mid-infrared photoacoustic spectrometer. In this case the MAE for the concentration predictions of ethanol and acetone are improved by 45 % and 20 % in comparison to purely synthetic training. This shows the capability of using synthetically generated data to train ANNs in combination with small amounts of measured data to further improve neural networks for gas sensing and the transferability between different sensing approaches.
The tensile as well as the fatigue behaviour of an experimental copper-alloyed metastable austenitic stainless steel has been investigated. Three batches were studied: (i) cast material as reference ...state and two states were recrystallised after (ii) rotary swaging and (iii) forward impact extrusion. Deformation-induced α’-martensite was formed during both rotary swaging and forward impact extrusion with volume fractions of 48 vol% and 4 vol%, respectively. The material was then heat treated to form a fine-grained, fully austenitic microstructure. These fine-grained material states exhibited a higher strength compared to the cast state. However, strain-hardening was highest in the cast state. The effect of the initial microstructure on the cyclic deformation behaviour and fatigue life was investigated using strain-controlled tests. During cyclic deformation all three states exhibited a martensitic phase transformation and concurrently a pronounced cyclic hardening with the formation of up to approximately 70 vol% α’-martensite. The fine-grained state produced by rotary swaging and reversion annealing showed a significantly higher fatigue life of about factor 4 compared to the coarse-grained cast state. Relative to forward impact extrusion, rotary swaging provided a higher fatigue life and fatigue strength.
Dual-frequency comb spectroscopy has emerged as a disruptive technique for measuring wide-spanning spectra with high resolution, yielding a particularly powerful technique for sensitive ...multi-component gas analysis. We present a spectrometer system based on dual electro-optical combs with subsequent conversion to the mid-infrared via tunable difference frequency generation, operating in the range from 3 to 4.7 \(\mu\)m. The simultaneously recorded bandwidth is up to 454(1) GHz and a signal-to-noise ratio of 7.3(2) x 10\(^2\) Hz\(^{-1/2}\) can be reached. The conversion preserves the coherence of the dual-comb within 3 s measurement time. Concentration measurements of 5 ppm methane at 3.3 \(\mu\)m, 100 ppm nitrous oxide at 3.9 \(\mu\)m and a mixture of 15 ppm carbon monoxide and 5 % carbon dioxide at 4.5 \(\mu\)m are presented with a relative precision of 1.4 % in average after 2 s measurement time. The noise-equivalent absorbance is determined to be less than 4.6(2) x 10\(^{-3}\) Hz\(^{-1/2}\).
In familial defective apolipoprotein B-100 (FDB) the presence of a mutant apolipoprotein (apo) B-100 (FDB3500Q/W) in LDL markedly reduces their affinity for the LDL receptor, leading to elevated LDL ...cholesterol levels. However, the hypercholesterolemia in most FDB patients is relatively mild when compared with, e.g., familial hypercholesterolemia (FH). In order to study mechanisms that may partly alleviate the clinical consequences of FDB, we investigated the in vivo kinetics of apoB-100-containing lipoproteins in five FDB heterozygotes (total cholesterol: 7.84 +/- 1.37 mmol/I; total apoB: 1.68 +/- 0.37 g/l; mean +/- SD) and six normolipidemic controls (4.61 +/- 0.62 mmol/l; 0.98 +/- 0.12 g/l) using a stable isotope approach. During and after a 10-12 h primed, constant infusion of either 13C6phenylalanine or 2H3leucine, tracer enrichment was determined in apoB-100 from ultracentrifugally isolated VLDL1 (Sf 60-400), VLDL2 (Sf 20-60), IDL (Sf 12-20), LDL1 (Sf 7-12), and LDL2 (Sf 0-7). The rates of apoB-100 production, catabolism, and transfer were estimated by model-based compartmental analysis. The overall fractional catabolic rate (FCR) of IDL apoB-100 in FDB was substantially increased (2.99 +/- 0.68 pools/day vs. 1.70 +/- 0.23 pools/day in controls, P < 0.01). The fractional rate of apoB-100 transfer from IDL to LDL in FDB was decreased (0.97 +/- 0.13 pools/day vs. 1.24 +/- 0.10 pools/day, P < 0.05). The FCR of LDL apoB-100 in FDB was decreased (0.18 +/- 0.07 pools/day vs. 0.56 +/- 0.05 pools/, P < 0.01). Finally, the input rate of LDL apoB-100 in FDB was markedly decreased (9.45 +/- 2.96 mg/kg day1 vs. 15.54 +/- 1.70 mg/kg day1, P < 0.05). Our data suggest that the relatively small increase of LDL concentrations in FDB is due to an increased clearance of LDL precursor particles via the LDL-receptor and apoE-receptors as well as a decreased conversion of IDL to LDL - two mechanisms that distinguish FDB from FH.