Modern predictive maintenance techniques have been significantly improved with the development of Industrial Internet of Things solutions which have enabled easier collection and analysis of various data. Artificial intelligence-based algorithms in combination with modular interconnected architecture of sensors, devices and servers, have resulted in the development of intelligent maintenance systems which outperform most traditional machine maintenance approaches. In this paper, a novel acoustic-based IoT system for condition detection of rotating machines is proposed. The IoT device designed for this purpose is mobile and inexpensive and the algorithm developed for condition detection consists of a combination of discrete wavelet transform and neural networks, while a genetic algorithm is used to tune the necessary hyperparameters. The performance of this system has been tested in a real industrial setting, on different rotating machines, in an environment with strong acoustic pollution. The results show high accuracy of the algorithm, with an average F1 score of around 0.99 with tuned hyperparameters.

Linear regression is a very simple machine learning model that is supposed to find linear relations between input and output data. Its use is limited since real-world random variables are almost never linearly correlated, but when there is limited amount of data for training, using other larger and more sophisticated models might be impossible. Even though simple linear relations are hard to come by in nature, more complex mathematical dependencies can often be linearized i.e. approximated using linear models. Additionally, by preprocessing the input data using nonlinear mathematical operations linear regression can artificially account for nonlinear relations. This methodology was tested on real data representing measurements in thermal power plant while the goal was predicting the specific energy of the coal that was being burnt in the plant. Estimating coals caloric value is of tremendous value since using this information power plant can be configured to run more efficiently.

Accurate and robust sensor measurements are essential for safe and efficient control in industrial systems. However, outliers in sensor data can produce unreliable estimates and lead to incorrect control or protection actions. To mitigate this risk, industrial plants often use two or more sensors to measure the same critical parameter. While this redundancy improves measurement reliability, it also requires a dedicated approach for combining multiple measurements into a single representative value. Existing methods for combined measurement derivation typically rely on static statistical measures, such as the mean or median, which lack adaptability in dynamic environments. This paper introduces a novel adaptive robust estimation algorithm that derives a single representative measurement from redundant sensor data. The approach combines maximum likelihood estimation (MLE) to assess the probability of contamination and to estimate the variances of both the true signal and outliers, with a dynamic switching mechanism that selects the mean or median based on current data characteristics. To enhance performance, the method integrates Kalman filtering for residual generation, thereby ensuring zero-mean inputs for robust estimation. The proposed methodology is evaluated using both simulated datasets and real-world measurements of negative gauge pressure from a thermal power plant. Experimental results show that the adaptive estimator significantly outperforms traditional static estimators in terms of accuracy and stability, particularly in the presence of sensor faults and outliers. This work advances the state of the art by providing a fully automated, real-time-capable solution for robust measurement fusion in safety-critical industrial applications.