Standard Absolute Error Research Articles

Fault Detection and Diagnosis for Liquid Rocket Engines Based on Long Short-Term Memory and Generative Adversarial Networks

The development of health monitoring technology for liquid rocket engines (LREs) can effectively improve the safety and reliability of launch vehicles, which has important theoretical and engineering significance. Therefore, we propose a fault detection and diagnosis (FDD) method for a large LOX/kerosene rocket engine based on long short-term memory (LSTM) and generative adversarial networks (GANs). Specifically, we first modeled a large LOX/kerosene rocket engine using MATLAB/Simulink and simulated the engine’s normal and fault operation states involving various startup and steady-state stages utilizing fault injection. Second, we created an LSTM-GAN model trained with normal operating data using LSTM as the generator and a multilayer perceptron (MLP) as the discriminator. Third, the test data were input into the discriminator to obtain the discrimination results and realize fault detection. Finally, the test data were input into the generator to obtain the predicted samples and calculate the absolute error between the predicted and the real value of each parameter. Then the fault diagnosis index, standardized absolute error (SAE), was constructed. SAE was analyzed to realize fault diagnosis. The simulated results highlight that the proposed method effectively detects faults in the startup and steady-state processes, and diagnoses the faults in the steady-state process without missing an alarm or being affected by false alarms. Compared with the conventional redline cut-off system (RCS), adaptive threshold algorithm (ATA), and support vector machine (SVM), the fault detection process of LSTM-GAN is more concise and more timely.

Local decadal prediction according to statistical/dynamical approaches.

Dynamical climate models present an initialization problem due to the poor availability of deep oceanic data, which is required for the model assimilation process. In this sense, teleconnection indices, defined from spatial and temporal patterns of climatic variables, are conceived as useful tools to complement them. In this work, the near‐term climate predictability of 35 temperature and 36 precipitation time series of three cities (Barcelona, Bristol and Lisbon) was analysed using two approaches: (a) a statistical–dynamical combination of self‐predictable teleconnection indices and long‐term climate projections on a local scale and (b) dynamical model outputs obtained from drift‐corrected decadal experiments. Fourier and wavelet analyses were used to assess the predictability of seven teleconnection indices thanks to a cross‐validation process (with differentiated training and validation periods). The standardized absolute error of teleconnection‐based prediction was compared with that obtained from a (9) multi‐model ensemble based on the Coupled Model Intercomparison Project Phase 5. Results showed that decadal predictions at horizons between 20 and 30 years are adequate for temperature and precipitation if a teleconnection‐based approach is used, while temperature is better predicted from a 5‐year horizon using drift‐corrected dynamical outputs.

Clean technology approach for the competitive binding of toxic metal ions onto MnO2 nano-bioextractant

The competitive extraction of Cr(III) onto Nauclea diderrichii seed epicarp doped with MnO2 nanoparticles (MnO2 nano-bioextractant (MNB)) in a single and binary batch system was studied. For validity of experimental data, chi square test, root mean square error, sum of the square errors, hybrid fractional error function, Marquart’s percent standard deviation and standard absolute error were used. Among the kinetic models used, pseudo-second-order and Langmuir equations gave the best fits for the experimental data, with $$q_{e}$$ (mg g) for the uptake of Cr(III) in single metal system onto MNB, then Cr(III) with Cd(II), Pb(II), Hg(II), KCl and CaCl2 in binary metal systems onto MNB were 2.611, then 1.989, 1.016, 2.208, 1.249 and 1.868 from kinetic standpoint, respectively. The initial sorption rates, h (mg/g/min), and half lives, $$t_{1/2}$$ (min), for the uptake of Cr(III) in single metal system onto MNB, then Cr(III) with Cd(II), Pb(II), Hg(II), KCl and CaCl2 in binary metal system onto MNB were 3.497, then 2.311, 2.274, 0.242, 2.956, 45.568 and 0.747, then 5.769, 1.766, 12.144, 1.762, and 2.415, respectively. Physicochemical surface analyses such as pH of point of zero charge, Brunauer–Emmett–Teller single point and multi-point techniques for surface area analyses, scanning electron microscopy and transmission electron microscopy were done on MNB and MnO2 nanoparticles in order to understand their surface microstructures. Desorption study showed that MNB can be recycled and used for future study. Hence, MNB showed good potential to remediate Cr(III) from wastewaters and polluted water.

Detection of growth-restricted fetuses in preeclampsia: a case-control study

Objective: To determine the diagnostic accuracy of detecting growth-restricted fetuses in women with and without preeclampsia. Methods: Over 2 years, parturients with reliable gestational ages, preeclampsia, and sonographic estimates of birth weights were matched (1:1) for gestational age with women without preeclampsia. Paired and unpaired t tests were used; P < .05 was significant. Odds ratios (ORs) and 95% confidence intervals (CIs) were calculated. Results: Two hundred eighty-seven preeclamptic women were identified and matched. In each group, mean (± standard deviation [SD]) gestational age was 34.9 ± 4.2 weeks, and 166 (57.8%) infants were born preterm. Fetal growth restriction (FGR) was significantly more common among women with preeclampsia (14.9%) than among controls (5.6%; OR 2.98, 95% CI 1.64, 5.44). The percentage of sonographic estimates within 10% of actual birth weight (57.5% versus 53.6%) was similar in the two groups (OR 1.16; 95% CI 0.84,1.62). Compared with normal growth, the mean (± SD) standardized absolute error was significantly higher among those with FGR regardless of group (preeclampsia 109 ± 100 versus 158 ± 152 g/kg; P = .009; control 117 ± 103 versus 233 ± 206 g/kg; P < .001). Fetal growth restriction was detected more commonly among preeclamptic women than among controls (11.6% versus 0%; OR 4.74 95% CI 0.25, 90.31). The sensitivity and positive predictive value of FGR detection were 10% and 50%, respectively, among women with preeclampsia and 0% each among controls. Conclusion: Although FGR was detected more frequently in fetuses of women with preeclampsia than in those of controls, the ability to predict it with sonography remained poor.

Detection of Growth-Restricted Fetuses in Preeclampsia

In Brief Objective To determine the diagnostic accuracy of detecting growth-restricted fetuses in women with and without preeclampsia. Methods Over 2 years, parturients with reliable gestational ages, preeclampsia, and sonographic estimates of birth weights were matched (1:1) for gestational age with women without preeclampsia. Paired and unpaired t tests were used; P < .05 was significant. Odds ratios (ORs) and 95% confidence intervals (CIs) were calculated. Results Two hundred eighty-seven preeclamptic women were identified and matched. In each group, mean (± standard deviation [SD]) gestational age was 34.9 ± 4.2 weeks, and 166 (57.8%) infants were born preterm. Fetal growth restriction (FGR) was significantly more common among women with preeclampsia (14.9%) than among controls (5.6%; OR 2.98, 95% CI 1.64, 5.44). The percentage of sonographic estimates within 10% of actual birth weight (57.5% versus 53.6%) was similar in the two groups (OR 1.16; 95% CI 0.84,1.62). Compared with normal growth, the mean (± SD) standardized absolute error was significantly higher among those with FGR regardless of group (preeclampsia 109 ± 100 versus 158 ± 152 g/kg; P = .009; control 117 ± 103 versus 233 ± 206 g/kg; P < .001). Fetal growth restriction was detected more commonly among preeclamptic women than among controls (11.6% versus 0%; OR 4.74 95% CI 0.25, 90.31). The sensitivity and positive predictive value of FGR detection were 10% and 50%, respectively, among women with preeclampsia and 0% each among controls. Conclusion Although FGR was detected more frequently in fetuses of women with preeclampsia than in those of controls, the ability to predict it with sonography remained poor. Antenatal detection of growth-restricted fetuses in preeclamptic women is poor.

Ultrasonographic estimate of birth weight at 24 to 34 weeks: A multicenter study

Objective: The study was intended to compare the accuracies of ultrasonographic estimates of birth weights among infants born between 24 and 34 weeks’ gestation at 3 tertiary centers. Study Design: In this retrospective study subjects were matched for gestational age (1:1); all underwent ultrasonographic examination within 2 weeks of delivery. The estimates of birth weight were obtained according to 26 published regression equations and their accuracies were assessed with the mean standardized absolute error. For each center the equation with the lowest error was selected to generate (1) receiver-operating characteristic curves for an estimate to identify actual weight <1500 g and (2) prediction limit calculations to determine the estimate that ensures at 70% confidence a birth weight >1500 g. Results: One hundred seventy-one cases were analyzed at each center. Comparison of the 26 mean standardized errors at each center indicated that (1) the range was rather wide (eg, 89 ± 87 to 365 ± 313 g/kg) and (2) 73% (19/26) of the equations had significantly (P < .05) different accuracies. Receiver-operator characteristic curves show that fetal weight estimates of ≥1600 g at 2 centers and ≥1700 g at the third center are required to predict actual birth weight <1500 g. Prediction limit calculation suggests that different fetal weight estimates (>1600 g at center I, >1900 g for the center II, and >1800 g at center III) are needed to predict actual weight > 1500 g with a 70% accuracy. Conclusions: Ultrasonographic estimates of weight for preterm infants, as obtained from 26 equations, are characterized by a rather wide range of accuracy; for most of the equations the accuracies of estimates differ markedly among centers. (Am J Obstet Gynecol 1998;179:909-16.)

Weight gain in pregnancy, triceps skinfold thickness, and blood pressure in offspring

To compare the accuracy of clinical and sonographic estimates of fetal weight made throughout the third trimester of pregnancy. Patients in early labor had fetal weight estimated by two approaches: 1) clinical evaluation and palpation followed by 2) sonographic mensuration of fetal biparietal diameter, abdominal circumference, and femur length applied to Hadlock's formula. The accuracy of these two methods of estimating fetal weight was compared using Student t test, Wilcoxon test, and chi2 tests. P < .05 was considered significant. Prediction limits (50th, 90th, and 95th percentiles) were calculated for both techniques by obtaining the range of actual weights associated for a particular estimated fetal weight (EFW). We enrolled 1034 parturients whose clinical EFWs yielded significantly higher mean (+/- standard deviation) simple error (48.2 +/- 411 g) and standardized absolute error (130 +/- 122 g/kg) than were obtained by use of sonographic formulas for EFW (-6.6 +/- 381 g and 104 +/- 89 g/kg, respectively). When the population was partitioned by gestational age, we found that sonographic EFW was more accurate than clinical EFW in preterm (n = 373) but not in term (n = 460) or post-term (n = 201) pregnancies. Prediction limits indicate that for a given EFW, for example, 800 g, the 90% ranges of actual weight based on clinical and sonographic EFW are 566-1829 g and 469-1667 g, respectively. The apparent superiority of sonographic EFW over clinical EFW applies principally to preterm pregnancies. The prediction limitation calculation suggests that either method, for any particular estimate between 500 and 4500 g, has limited value in the estimation of actual birth weight, because this outcome is highly variable and frequently lies outside of the useful bandwidth (+/- 10%) for prospective management.

Intrapartum detection of a macrosomic fetus: clinical versus 8 sonographic models.

The purpose of this study was to determine whether clinical or sonographic models have 1) the highest accuracy in differentiating newborns with birth-weights > or = 4,000 g (macrosomia) versus < or = 3,999 g, and 2) among macrosomics which method of predicting birth-weight has the lowest percentage error. Prospectively, 602 consecutive parturients at term had a clinical estimate of birth-weight followed by sonographic measurement of fetal parts. The sonographic prediction of birth-weight was derived using 8 different models that utilize either 1 measurement or a combination of 2 to 4 parameters. The incidence of macrosomia was 11.1% (67 of 602). Analysis of ROC curves indicated that clinical predictions (w = 0.85) were significantly better than 4 of the 8 sonographic models. The mean standardized absolute error among macrosomic newborns is significantly lower when predictions are derived clinically (99 +/- 70 g/kg) than using 1 or 2 fetal parts. Sonographic assessment of birth-weight is not significantly more accurate in the detection of a macrosomic fetus than clinical predictions.

Sonographic assessment of birth weight among breech presentations

The objective in this study was to determine whether there was a difference in the accuracy of sonographically determined birth weight estimate in fetuses presenting by the breech as compared to the vertex. Prospectively, 104 consecutive singleton fetuses presented by the breech presentation were identified who also had a sonographic estimate of birth weight within 72 h of delivery. The control group (n = 104) consisted of fetuses presenting by the vertex of women who had the following hospital number and who were of similar demographic characteristics; the birth weight estimate in the control group had to be within 100 g of that of the women with the breech fetus. The estimate of the birth weight was derived for both groups using seven published models. Prediction limits were calculated for an estimated vs. actual birth weight for parturients with breech presentation. Among the study group, the mean standardized absolute error was similar for the seven models (range 120 +/- 104 to 138 +/- 118 g/kg; p = 0.08). For six of the models, the mean standardized absolute error was significantly higher for fetuses in the breech presentation than for fetuses in vertex presentation (range 80 +/- 63 to 112 +/- 95 g/kg; p = 0.02 to < 0.0001). Prediction limit calculations indicate that 99% of the time the actual birth weight was between 1500 and 3999 g if the estimate of birth weight was 2400-2900 g. We conclude that, for fetuses in a breech presentation, the error with most models is significantly higher than for a singleton in the vertex position.

Estimators of birth weight in pregnant women requiring insulin: A comparison of seven sonographic models

To determine if the relative accuracy of the sonographic estimate of birth weight among diabetic gravidas requiring insulin improves significantly as more fetal measurements are used. We studied 172 diabetic women requiring insulin who had sonographic measurements of fetal parts within 7 days of delivery. Friedman nonparametric analysis of variance followed by Dunn multiple comparison and chi 2 were used to assess the relative accuracy of the seven models. Prediction limits were calculated to determine the estimate of fetal weight that would ensure (with 90% accuracy) that the newborn was macrosomic (at least 4 kg). The mean (+/- standard deviation [SD]) birth weight was 3388 +/- 727 g, and the frequency of macrosomia at term gestation was 19.4% (29 of 149). The mean standardized absolute error (g/kg) based on abdominal circumference (AC) and femur length (FL) (86 +/- 72 g/kg) was not significantly different from the other models (range 84 +/- 72 to 116 +/- 99 g/kg, P > .05). The percent of estimate within 10% of actual birth weight using AC and FL (65%) was similar to the other models (53.4-66.2%). Regardless of the White classification, the estimation of fetal weight using AC and FL had an accuracy similar to the other six regression equations. To ensure that the birth weight is at most 3999 g, the estimated fetal weight should be 3200 g, and, conversely, if the predicted weight is 4700 g, then the newborn is macrosomic. Among patients requiring insulin, estimation of birth weight using AC and FL is as accurate as more complicated models. The ability to detect macrosomia by ultrasound is limited.

Intrapartum Assessment by House Staff of Birth Weight Among Twins

To determine among twins in labor: 1) the relative accuracy of an intrapartum sonographic estimate of the birth weight for both fetuses using biparietal diameter and abdominal circumference, 2) the accuracy of detecting discordant growth (difference in actual birth weights greater than or equal to 15%), and 3) the estimate of fetal weight for nonvertex twin B that would reliably avoid breech extraction of infants less than 1500 g. Retrospectively, we identified and analyzed parturients with twins who had an intrapartum sonogram performed by a house officer assigned to the labor and delivery suite. The mean birth weight (+/- standard deviation) for the twin A group was 1910 +/- 628 g and for twin B was 1869 +/- 668 g. The mean standardized absolute errors for the twin A group (121 +/- 118 g/kg) and the twin B group (92 +/- 67 g/kg) were not significantly different (P = .06). Analysis of variance revealed that regardless of the presentation of the fetuses, the mean standardized absolute error was not significantly different (P = .10). Using a difference in the estimates of birth weight of 15% or greater, the positive and negative predictive values of detecting discordant growth within a twin pair were 53 and 83%, respectively. Among 30 vertex-nonvertex twin pairs, 12 of the second fetuses had actual birth weights of 1500 g or less, and all were estimated to weigh less than 1700 g. The intrapartum sonographic estimate of fetal weight in twin pregnancy by house staff appears reliable, and the accuracy of prediction is similar regardless of presentation, discordance, or actual birth weight greater or less than 1500 g. To avoid vaginal delivery of a persistent nonvertex twin B with a birth weight of 1500 g or less, a sonographic estimate of 1700 g for the second fetus may be adequate.

Ultrasound Fetal Weight Estimation: Precision or Guess Work?

We compared the prediction of birth-weight based on factors shown by epidemiological studies to influence it with the performance of 2 clinical examples of ultrasonic fetal weight estimation reported in the literature. The standardized absolute error of the 2 examples of ultrasonic fetal weight estimation was similar to estimation by multiple regression of birth-weight on maternal age, parity, height and weight in 11,516 Chinese mothers. This severely limits the clinical utility of the estimation derived from these ultrasonic measurements.