It is generally known that QoE reflects the actual customer experience of network services. This parameter is not limited to network performance, but also includes aspects of customer behavior, ISP brand reputation, service standards and cost. That is why, in order to meet the needs of customers, providers should regularly determine the current quality of service to subscribers, which definitely correlates with QoE. To establish these relationships, it was decided to use machine learning models. In this regard, the paper recommends applying a machine learning model called Random Forest for data analysis, where the stages of analysis are considered in detail, taking into account historical information about QoS and QoE, respectively. In the course of the study, it was possible to test the method of QoE prediction based on QoS parameters based on machine learning algorithms. It was established that the applied methods allow to reduce the amount of data without losing the accuracy and efficiency of the calculations performed by the machine learning model. Thus, the presented approach can be applied by telecommunication providers and mobile operators in their activities for continuous monitoring of user satisfaction with the provided services.
In this paper, we introduce a scalable, AI-driven chatbot designed for real-time diagnostics in Industry 4.0 manufacturing plants. By incorporating Google Dialogflow and a plant-specific knowledge graph, the chatbot enables seamless interaction with users while requiring minimal reconfiguration when adapting to new environments. The integration of BERT in the back-end empowers the chatbot to efficiently identify errors. Furthermore, a self-learning module continuously scans the manufacturing execution system (MES) database for new errors, generating learning data for the automatic re-training of the BERT-based neural network. Although there are opportunities for refinement, the chatbot exhibits significant potential as a scalable, supplementary first-line support system in the manufacturing sector.
Food safety and quality are extremely important for consumer satisfaction. Eggs are an important source of protein worldwide and occupy an important place in many industries. Therefore, detecting and preventing damage that may occur in the egg production process is of great importance in terms of both protecting the health of the consumer and reducing the costs in the egg industry. However, the suitability and quality of eggs for consumption depends on various external factors throughout their journey. Among these factors, physical damage is one of the most common and important. In the egg industry, damage to eggs and therefore not being able to reach the consumer is an important cost factor. This means a great loss, especially for businesses that produce on an industrial scale. Damage detection and prevention methods can help reduce these losses. In this study, a dataset containing a total of 569 images in three classes was used to detect egg damage. There are three classes in the dataset named crack, empty and good. For the classification of images, VGG16 and VGG19 models from CNN (Convolutional Neural Network) architectures were used with transfer learning method. As a result of the classification, 93.8% classification success was obtained from the VGG16 model and 95.1% from the VGG19 model. F1 Score, precision and recall metrics of the classification performance of the model were also analyzed. By considering the advantages and disadvantages of detecting egg damage in the image, a basis for future studies in this field has been established.
An optimal control system for the production of mineral fertilizers in a fluidized bed granulator has been built. The main control channel is the dependence of the coolant temperature and the temperature of the granules. The criterion of optimality for the control system is obtained. The transfer function was obtained using the MATLAB System Identification package. The transition to the vector-matrix model is made. The Riccati equation is solved. The main factors influencing the process are identified and a control system with LQR-regulator is created, which meets the requirements of stability, dynamics and reliability. A transient process is obtained, which brings the temperature of the granules to a given level in 364 K. The transition process with LQR-regulator lasts 18.7 seconds, which is a real value for regulators of this type.
In this investigation, we present a new method of Author Identification (AI) based on two different modalities: Transcript Text (TT) and Speech Signal (SS). This research work is performed on a corpus of several discourses given by Famous Politicians in English (FPOL corpus). The originality of this work is that we use both Speech Signal and its Text Transcript during the classification process. That is, 14 famous politicians are considered, where 6 are female politicians and 8 are male politicians. The first analysis uses the text modality based on three classifiers and three features: Character 3Grams, Character 4Grams and Words, while the second analysis is based on those textual features in association with the average Pitch (Fo) of the speech signal modality. The proposed association uses a logical filtering based on the average Pitch range in order to correct the errors of identification if any. The experiments of author identification conducted on the FPOL corpus show that the association of the two modalities enhances considerably the identification performances. The average accuracy improvement brought by the speech modality is about 5% for character 3Grams and 4Grams, and it is about 12% of improvement for words.
