Automatic Speech Recognition can be considered as a transcription of spoken utterances into text which can be used to monitor/command a specific system. In this paper, we propose a general end-to-end approach to sequence learning that uses Long Short-Term Memory (LSTM) to deal with the non-uniform sequence length of the speech utterances. The neural architecture can recognize the Arabic spoken digit spelling of an isolated Arabic word using a classification methodology, with the aim to enable natural human-machine interaction. The proposed system consists to, first, extract the relevant features from the input speech signal using Mel Frequency Cepstral Coefficients (MFCC) and then these features are processed by a deep neural network able to deal with the non uniformity of the sequences length. A recurrent LSTM or GRU architecture is used to encode sequences of MFCC features as a fixed size vector that will feed a multilayer perceptron network to perform the classification. The whole neural network classifier is trained in an end-to-end manner. The proposed system outperforms by a large gap the previous published results on the same database.

Efficient routing and scheduling of vehicles has significant economic implications for both the public and private sectors. To this purpose, we propose in this study a decision support system which aims to optimize the classical Capacitated Vehicle Routing Problem by considering the existence of different vehicle types (with distinct capacities and costs) and multiple available depots, that we call the Multi-Depot Heterogeneous Vehicle Routing Problem with Time Window (MDHVRPTW) by respecting a set of criteria including: schedules requests from clients, the heterogeneous capacity of vehicles..., and we solve this problem by proposing a new scheme based on the application of the bio-inspired genetic algorithm heuristics and by embedding a clustering algorithm within a VRPTW optimization frame work, that we will specify later. Computational experiments with the benchmark test instances confirm that our approach produces acceptable quality solutions compared with the best previous results in similar problems in terms of generated solutions and processing time. Experimental results prove that our proposed genetic algorithm is effective in solving the MDHVRPTW problem and hence has a great potential.

This paper is concerned with fault detection and diagnosis problem in manufacturing systems. In such industrial environment, production systems are subject to several faults caused by a number of factors including the environment, the accumulated wearing, usage, etc. However, due to the lack of accuracy or fluctuation of data, it is oftentimes impossible to evaluate precisely the correct classification rate of faults. In order to classify each type of fault, neural networks and fuzzy logic are two different intelligent diagnosis methods that are more applied now, and each has its own advantages and disadvantages. A new hybrid fault diagnosis approach is introduced in this paper that considers the combined learning algorithm and knowledge base (Fuzzy rules) to handle ambiguous and even erroneous information. Therefore, to enhance the classification accuracy, three perceptron models including: linear perceptron (LP), multilayer perceptron (MLP) and fuzzy perceptron (FP) have been respectively established and compared. The conditional risk function "PDF" that measures the expectation of loss when taking an action is presented at the same time. We evaluate the proposed hybrid approach "Variable Learning Rate Gradient Descent with Bayes' Maximum Likelihood formula" VLRGD-BML on dataset of milk pasteurization process and compare our approach with other similar published works for fault diagnosis in the literature. Comparative results indicate the higher efficiency and effectiveness of the proposed approach with fuzzy perceptron for uncertain fault diagnosis problem.