Model order reduction (MOR) methods enable the generation of real-time-capable digital twins, with the potential to unlock various novel value streams in industry. While traditional projection-based methods are robust and accurate for linear problems, incorporating machine learning to deal with nonlinearity becomes a new choice for reducing complex problems. These kinds of methods are independent to the numerical solver for the full order model and keep the nonintrusiveness of the whole workflow. Such methods usually consist of two steps. The first step is the dimension reduction by a projection-based method, and the second is the model reconstruction by a neural network (NN). In this work, we apply some modifications for both steps respectively and investigate how they are impacted by testing with three different simulation models. In all cases Proper orthogonal decomposition is used for dimension reduction. For this step, the effects of generating the snapshot database with constant input parameters is compared with time-dependent input parameters. For the model reconstruction step, three types of NN architectures are compared: multilayer perceptron (MLP), explicit Euler NN (EENN), and Runge–Kutta NN (RKNN). The MLPs learn the system state directly, whereas EENNs and RKNNs learn the derivative of system state and predict the new state as a numerical integrator. In the tests, RKNNs show their advantage as the network architecture informed by higher-order numerical strategy.

There is a fundamental concern regarding the prediction of kiwifruit yield based on the concentration of nutrients in the leaf (2–3 months before fruits harvesting). For this purpose, the current study was designed to employ an artificial neural network (ANN) to evaluate the kiwi yield of Hayward cultivar. In this regard, 31 kiwi orchards (6–7 years old) in different parts of Rudsar, Guilan Province, Iran, with 101 plots (three trees in every plot) were selected. The complete leaves of branches with fruits were harvested, and the concentration of nitrogen, potassium, calcium, and magnesium measured. After fruit harvesting in late November, the fruit yield of each plot was evaluated along with the fresh and dry weights of the fruit. The ANN analyses were carried out using a multi-layer perceptron with the Langburge-Marquardt training algorithm. Using calcium (Ca) as input data (Ca-model) was more accurate than using nitrogen (N-model). The maximum R2 and the lowest root mean square error was obtained when all nutrients and related ratios were considered as input variables. Since the difference between the proposed model and the model fitted by the calcium variable (Ca-model) was only about 6%, the Ca-model is recommended.

The performance of Global Navigation Satellite System (GNSS) and Micro-Electro-Mechanical System (MEMS)-based Inertial Navigation System (INS) integrated navigation is reduced during GNSS outages. To bridge the period during GNSS outages, a novel hybrid intelligent algorithm incorporating a Discrete Grey Predictor (DGP) and a Multilayer Perceptron (MLP) neural network (DGP-MLP) is proposed. The DGP-MLP is used to provide a pseudo-GNSS position to correct the INS errors during GNSS outages; the DGP uses the GNSS position information of the latest few moments to predict the position of future moments; in the process of DGP-MLP, the MLP is used to modify the prediction errors of DGP, and the MLP is improved by adding momentum terms and adaptively adjusting the learning rate and momentum factor. To evaluate the effectiveness of the proposed methodology, four GNSS outages in different cases over a real field test data were employed. The experimental results demonstrate that the proposed methodology can significantly improve positioning accuracy during GNSS outages.

Several researchers have developed prediction equations to estimate the metabolisable energy (ME) of energetic and protein concentrate feedstuffs used in diets for broilers. The ME is estimated by considering CP, ether extract, ash and fibre contents. However, the results obtained using traditional regression analysis methods have been inconsistent and new techniques can be used to obtain better estimate of the feedstuffs’ energy value. The objective of this paper was to implement a multilayer perceptron network to estimate the nitrogen-corrected metabolisable energy (AMEn) values of the energetic and protein concentrate feeds, generally used by the poultry feed industry. The concentrate feeds were from plant origin. The dataset contains 568 experimental results, all from Brazil. This dataset was separated into two parts: one part with 454 data, which was used to train, and the other one with 114 data, which was used to evaluate the accuracy of each implemented network. The accuracy of the models was evaluated on the basis of their values of mean squared error, R2, mean absolute deviation, mean absolute percentage error and bias. The 7-5-3-1 model presented the highest accuracy of prediction. It was developed an Excel® AMEn calculator by using the best model, which provides a rapid and efficient way to predict the AMEn values of concentrate feedstuffs for broilers.

Two types of artificial neural networks, multilayer perceptron (MLP) and self-organizing feature map (SOM) were used to detect mastitis by automatic milking systems (AMS) using a new mastitis indicator that combined two previously reported indicators based on higher electrical conductivity (EC) and lower quarter yield (QY). Four MLPs with four combinations of inputs were developed to detect infected quarters. One input combination involved principal components (PC) adopted for addressing multi-collinearity in the data. The PC-based MLP model was superior to other non-PC-based models in terms of less complexity and higher predictive accuracy. The overall correct classification rate (CCR), sensitivity and specificity of this model were 90·74%, 86·90% and 91·36%, respectively. The SOM detected the stage of progression of mastitis in a quarter within the mastitis spectrum and revealed that quarters form three clusters: healthy, moderately ill and severely ill. The clusters were validated using k-means clustering, ANOVA and least significant difference. Clusters reflected the characteristics of healthy and subclinical and clinical mastitis, respectively. We conclude that the PC based model based on EC and QY can be used in AMS to detect mastitis with high accuracy and that the SOM model can be used to monitor the health status of the herd for early intervention and possible treatment.

This article describes the use of an electronic nose (the Cyranose 320) to sense and identify three species of bacteria responsible for ear, nose, and throat (ENT) infections. Gathered data were a complex mixture of different chemical compounds. An innovative approach for classifying the bacteria data was investigated by using a combination of several clustering algorithms. The best results suggest that the three classes of bacteria examined can be predicted with up to 98% accuracy, allowing more precise diagnosis of ENT infection in patients. This type of bacteria data analysis and feature extraction is difficult, but it can be concluded that combined use of the analysis methods described here can solve the feature extraction problem with very complex data and enhance the performance of electronic noses.

This article presents the results of research concerning possibilities of applying multilayer perceptron type of neural network for fault diagnosis, state estimation, and prediction in the gas pipeline transmission network. The influence of several factors on accuracy of the multilayer perceptron was considered. The emphasis was put on the multilayer perceptrons' function as a state estimator. The choice of the most informative features, the amount and sampling period of training data sets, as well as different configurations of multilayer perceptrons were analyzed.