Daniel Zurita

Intelligent sensor based on acoustic emission analysis applied to gear fault diagnosis

The development of intelligent and autonomous monitoring systems applied to rotating machinery, represents the evolution towards the automatic industrial plants supervision. In this regard, an acoustic emission based intelligent sensor is presented in this work. The proposed sensor records regularly the acoustic emission signal generated by gearboxes. A time domain statistical analysis is applied in order to characterize the acquired data. Afterwards, a neural network based algorithm is applied to detect gear fault patterns. Finally, the diagnosis result is sent through a wireless transceiver to the central control unit. Moreover, in order to reach a real autonomous operation, the sensor power is approached by different energy harvesting solutions.

Industrial Time Series Modelling by Means of the Neo-Fuzzy Neuron

Industrial process monitoring and modeling represent a critical step in order to achieve the paradigm of zero defect manufacturing. The aim of this paper is to introduce the neo-fuzzy neuron method to be applied in industrial time series modeling. Its open structure and input independence provide fast learning and convergence capabilities, while assuring a proper accuracy and generalization in the modeled output. First, the auxiliary signals in the database are analyzed in order to find correlations with the target signal. Second, the neo-fuzzy neuron is configured and trained accordingly by means of the auxiliary signal, past instants, and dynamics information of the target signal. The proposed method is validated by means of real data from a Spanish copper rod industrial plant, in which a critical signal regarding copper refrigeration process is modeled. The obtained results indicate the suitability of the neo-fuzzy neuron method for industrial process modeling.

Enhanced Industrial Machinery Condition Monitoring Methodology Based on Novelty Detection and Multi-Modal Analysis

This paper presents a condition-based monitoring methodology based on novelty detection applied to industrial machinery. The proposed approach includes both the classical classification of multiple a priori known scenarios, and the innovative detection capability of new operating modes not previously available. The development of condition-based monitoring methodologies considering the isolation capabilities of unexpected scenarios represents, nowadays, a trending topic able to answer the demanding requirements of the future industrial processes monitoring systems. First, the method is based on the temporal segmentation of the available physical magnitudes, and the estimation of a set of time-based statistical features. Then, a double feature reduction stage based on principal component analysis and linear discriminant analysis is applied in order to optimize the classification and novelty detection performances. The posterior combination of a feed-forward neural network and one-class support vector machine allows the proper interpretation of known and unknown operating conditions. The effectiveness of this novel condition monitoring scheme has been verified by experimental results obtained from an automotive industry machine.

Enhanced load forecasting methodology by means of probabilistic prediction intervals estimation

The improvement of the forecasting accuracy for prediction of future loads has been object of exhaustive study in the recent years, to the point that a wide variety of methodologies which have been proved to be valid and practical exists. However, most methodologies for demand forecasting do not handle uncertainties of the resulting model, which leads to a nonproper interpretation of the forecasted outcomes. In this context, this work presents a novel load forecasting methodology in order to quantify the model uncertainties and complement the resulting information by means of adaptive confidence intervals. First, an input selection technique based on Genetic Algorithms is used to select the best combination of inputs in order to obtain a state-of-the-art model by means of Adaptive Neuro-Fuzzy Inference Systems. Then the data space is analyzed in terms of error probability of the model outcomes. The principal component analysis is used to visualize the error probability in a 2-D map. Finally, an Artificial Neural Network is used to perform the identification of the error probability associated to new measurements. In conjunction with the forecasting model, the proposed classifier extends the resulting information with an adaptive confidence intervals and its probability distribution. The effectiveness of this enhanced load forecasting methodology has been verified by experimental data obtained from an automotive plant.

Remaining useful life estimation of ball bearings by means of monotonic score calibration

The estimation of remaining useful life applied to industrial machinery and its components is one of the current trends in the advanced manufacturing field. In this context, this work presents a reliable methodology applied to ball bearings health monitoring. First, the proposed methodology analyses the available vibration and temperature data by means of the Spearman coefficient. This step allows the identification of the most significant monotonic relationship between features and the evolution of the remaining useful life. The method is complemented by means of the application of one-class support vector machine in order to obtain the remaining useful life indication trough the mapping of the classification scores. The proposed scheme shows a significant accuracy and reliability of the degradation detection due to the coherent management of the information. This fact is experimentally demonstrated by a run-to-failure test bench and the comparison with classical approaches.

Distributed neuro-fuzzy feature forecasting approach for condition monitoring

The industrial machinery reliability represents a critical factor in order to assure the proper operation of the whole productive process. In regard with this, diagnosis schemes based on physical magnitudes acquisition, features calculation, features reduction and classification are being applied. However, in this paper, in order to enhance the condition monitoring capabilities, a forecasting approach is proposed, in which not only the current status of the system under monitoring in identified, diagnosis, but also the future condition is assessed, prognosis. The novelties of the proposed methodology are based on a distributed features forecasting approach by means of adaptive neuro-fuzzy inference system models. The proposed method is validated by means of an accelerated bearing degradation experimental platform.

Time series forecasting by means of SOM aided Fuzzy Inference Systems

The forecast of industrial process time series represents a critical factor in order to assure a proper operation of the whole manufacturing chain, as it allows to act against any process deviation before it affects the final manufactured product. In this paper, in order to take advantage from process relations and improve forecasting performance, a prediction method based in Adaptive Neuro Fuzzy Inference System (ANFIS) and Self-Organizing Maps is presented. The novelties of the proposed method are based on considering, as an input of an ANFIS model, the interrelations of process variables regarding the signal that wants to be forecasted, by means of topology preservation SOM. An experimental study performed with real industrial data from a cooper manufacturing plant indicated the suitability of the proposed method in time series forecasting applications.

Novelty detection methodology based on multi-modal one-class support vector machine

The lack of information of complicated industrial systems represents one of the main limitation to implement condition monitoring and diagnosis systems. Novelty detection framework plays an essential role for monitoring systems in which the information about the different operation conditions or fault scenarios is unavailable or limited. In this context, this work presents a novelty detection approach applied to a main rotatory element of an industrial packaging machine, a camshaft. The developed novelty detection method begins with the assumption that only data corresponding to a healthy operation of the machine is available, and the objective is to detect anomalies in the behavior of the machine. To monitor the packing machine, first, the current signals acquired from the main motor are processed by means of a normalized time-frequency map. Next, a set of features are calculated from the frequency maps. Then a set of novelty models are trained. When abnormal data is detected, an alarm will be activated to be confirmed by the user. The proposed methodology includes the re-training of the novelty detection models to include such behaviors. The proposed methodology shows a good performance to identify abnormal behavior on the machine and successfully incorporate novel scenarios.

Diagnosis method based on topology codification and neural network applied to an industrial camshaft

Since the last years, there is an increasing interest from the industrial sector to provide the electromechanical systems with diagnosis capabilities. In this context, this work presents a novel monitoring scheme applied to diagnose faults in the main rotatory element of an industrial packaging machine, the camshaft. The developed diagnosis method considers a coherent procedure to process the acquired measurement. First, the current signals acquired from the main motor are processed in a normalized time-frequency map. Next, the characteristics fault patterns are identified and numerically characterized. A double self-organized map structure is proposed to manage the information till compress it to just two features by means of a topology codification of the data space. Finally, a neural network based classification algorithm is used to classify the condition of the camshaft. The effectiveness of this condition monitoring scheme has been verified by experimental results obtained from industrial machinery.

Hierarchical classification scheme based on identification, isolation and analysis of conflictive regions

A great deal of effort is being made to increase accuracy and reliability of Condition Based Maintenance systems; for instance, by improved feature selection strategies or optimization approaches of classifier parameters. In this work a novel classification methodology is presented, covering from the characterization of the acquired physical magnitudes to the configuration of the classification algorithms. The proposed methodology provides a more accurate classification structure by identifying and isolating conflictive regions in the classification space and by specialized feature reduction and classification stages for them. The proposed Hierarchical Classification Scheme is composed by sequential layers, in which the clear membership regions are identified first, and the conflictive regions of classification are tackled in upper levels. Such treatment of the conflictive regions is based on new feature space transformation to provide an optimized data understanding and, then, better chances of classification. Improving classification with this method compared to other alternatives implies the avoidance of over-fitting the classification training. Also, the proposed methodology, due to its hierarchical structure nature, offers the opportunity to configure the feature reduction and classification algorithms to obtain the optimal data management.