Helio Fiori de Castro

Development of hybrid algorithm based on simulated annealing and genetic algorithm to reliability redundancy optimization

Purpose – The purpose of this paper is to present an application of the simulated annealing algorithm to the redundant system reliability optimization. Its main aim is to analyze and compare this optimization method performance with those of similar application.Design/methodology/approach – The methods that were used to compare results are the genetic algorithm, the Lagrange Multipliers, and the evolution strategy. A hybrid algorithm composed by simulated annealing and genetic algorithm was developed in order to achieve the general applicability of the methods. The hybrid algorithm also tries to exploit the positive aspects of each method.Findings – The results presented by the simulated annealing and the hybrid algorithm are significant, and validate the methods as a robust tool for parameter optimization in mechanical projects development.Originality/value – The main objective is to propose a method for redundancy optimization in mechanical systems, which are not as large as electric and electronic syst...

Convergence on unbalance identification process based on genetic algorithm applied to rotating machinery

This work focusses on the application of genetic algorithm to fault detection in rotating systems and an uncertainty analyses were applied to mass unbalance identification. A finite element model is used to represent the system. The journal bearing stiffness and damping coefficients have already been identified by a model updating process, and they have been included in the model of the system. In order to identify the unknown fault parameters, a genetic algorithm search process is applied. The convergence of the process was investigated with regard to the influence of the main genetic algorithm parameters, taking into account the mean and standard deviation related to the objective function convergence in several cases.

An analysis on the supporting structure representative model in rotating systems

Power generation systems are composed by several rotating system, which are supported by bearings, and are installed on foundation structure. For this reason, a representative model needs to take into account the components effects. Rotating systems numerical models are well known by the scientific community specialized in rotor-dynamics to predict the system dynamic behaviour. Instead, the foundation numerical model is more sensitive to uncertainties. Experimental models can be the solution to the foundation representation. In this case, some difficulty in identifying the modal parameters, mainly the damping factors, can influence the results. This work proposes a comparison between complete system response using both experimental models of mechanical impedance and mixed coordinates to represent a test-rig foundation. A classical approach is the assembly of the impedance matrix of the supporting structure directly from the flexibility matrix inversion. However, this technique can be limited by the number of degrees of freedom associated to the rotor-structure connecting elements. Therefore, a solution based on modal parameters of mass, stiffness and damping to represent the foundation is also applied and both models are associated to a rotor-bearing system model for comparison.

Experimental Performance Evaluation of Magnetic Actuator used in Rotating Machinery Analysis

A magnetic actuator as excitation source in rotative systems is proposed, in order to accomplish modal analysis without contact between the actuator and the rotor. Although the use of electromagnets for applying forces to rotating machinery have been carried out with high performance level (for example, magnetic bearings), the development of a conveniently easy and cheap device for laboratory application presents interesting contribution to experimental methods used in test rigs based on similarity design to rotating machinery. The initial concept of the magnetic actuator proposed here is simple, but enables either the external excitation without contact or the vibration control when associated with a controller system. However, the calibrations of performance characteristics to attend the dynamic demand of the system in not so trivial. Following this focus, the paper brings practical experience and discussion about the development, calibration and performance analysis of a magnetic actuator used for rotating machinery tests. The influence of the electrical current in the actuator coils, the air-gap between actuator and rotative system, the type of surface of the actuator poles (flat or curved) as well as excitation frequency was experimentally verified and compared with theoretical concepts. Force estimation was carried out and compared with the measured force. The estimation was based on the magnetic flux density, measured by hall sensors, or input current and initial air-gap.

Magnetic Actuator Modelling for Rotating Machinery Analysis

Rotating machines have a wide range of application such as airplanes, factories, laboratories and power plants. Lately, with computer aid design, shafts finite element models including bearings, discs, seals and couplings have been developed, allowing the prediction of the machine behavior. In order to keep confidence during operation, it is necessary to monitor these systems, trying to predict future failures. One of the most applied technique for this purpose is the modal analysis. It consists of applying a perturbation force into the system and then to measure its response. However, there is a difficulty that brings limitations to the excitation of systems with rotating shafts when using impact hammers or shakers, once due to friction, undesired tangential forces and noise can be present in the measurements. Therefore, the study of a non-contact technique of external excitation becomes of high interest. In this sense, the present work deals with the study and development of a finite element model for rotating machines using a magnetic actuator as an external excitation source. This work also brings numerical simulations where the magnetic actuator was used to obtain the frequency response function of the rotating system.

Application of Bayesian Inference to Unbalance Identification in Rotors

This work proposes the use of Bayesian inference to fault parameters identification taking into account the stochastic characteristic of the system. The objective is to estimate the unbalance parameters, as the unbalance moment, phase angle and axial position of the unbalance force applied to the rotor. Therefore, experimental tests with the rotor to obtain the unbalance response are performed. The statistical distribution of each parameter is obtained using a Markov Chain Monte Carlo method (MCMC), simulated with the Delayed Rejection Adaptive Metropolis algorithm (DRAM). Thus, the residual between experimental and numerical response is calculated and applied to a Bayesian inference analysis, obtaining information about the unbalance parameters, which are summarized in statistics for each parameter distributions.

Assessment of Numerical Model to Determination of Parameters to Stability Experimental Tests

Rotating machines have to meet rigorous requirements in order to prevent instability during its operation. In the context of stability analysis, experimental tests such as stepped sine are widely used to determine the modal parameters of rotating systems: forward and backward natural frequencies and damping factors. Nevertheless, classical modal analysis techniques require prior knowledge of the system behavior, so that rotational speed and external excitation frequencies can be defined for the experimental tests. This work aims the assessment of model based numerical calculations to reduce or even stave off the preliminary tests. Validation starts with the evaluation of lubricated bearings model by shaft center locus. Afterwards, mass unbalance response is evaluated, and then, the stability analysis is conducted based on the logarithmic decrement. Finally, the numerical evaluation is compared to an experimental procedure regarding the precision of predicted critical frequencies for the tests and the evaluation of stability threshold.

Multi-objective optimization in redundant system considering load sharing

Abstract Reliability allocation problem (RAP) deals with the dilemma between reliability (or availability) increase and some undesirable consequences, as an increase in cost. Therefore, new installations or maintenance investments should be optimally allocated to maximize the reliability (or availability), considering all restrictions. This paper proposes a solution for the RAP problem, taking into account failure dependency among redundant components. Because of that, a stochastic approach, based on Markov Chain, is applied. The multi-objective problem is solved with NSGA-II (Non-Dominated Sorting Genetic Algorithm II), assuming availability and overall costs as objective functions. Two examples are tested with the proposed approach. The first case is a hypothetical system with 5 subsystems in series. The second example is a real industrial application. In both cases, a Pareto front is obtained, allowing an analysis of viable solutions to be adopted. Additionally, the failure dependency effect is also present in the optimization results. Therefore, it is possible to investigate the influence on the system availability and final cost.

Stochastic Collocation Approach to Bayesian Inference Applied to Rotating System Parameter Identification

The analyzed problem is the identification of fault parameters taking into account the stochastic characteristic of the system. The objective is to estimate the unbalance parameters, as the unbalance moment, phase angle and axial position of the unbalance force applied to the rotor. Therefore, experimental tests with the rotor to obtain the unbalance response is performed. This work aims the comparison between Bayesian inference with Markov Chain Monte Carlo method (MCMC), using Delayed Rejection Adaptive Metropolis algorithm (DRAM), and Stochastic Collocation through Generalized polynomial chaos expansion. This method has computational cost smaller than the MCMC methods, and it could be used as an alternative method for stochastic simulation. The Bayesian inference with MCMC and DRAM is based on previous works. However, the application of the MCMC have a high computational cost. Therefore, the Stochastic collocation is introduced into the likelihood function of the Bayes theorem for a faster convergence rate. The low computational cost of the collocation is evaluated and the results of both methods are compared to determine the convergence and precision of the collocation method.

Application of Stochastic Collocation on Eigenfrequencies Analysis of a Rotor-Bearing System

Rotating machines have a remarkable importance on industry and understanding their behavior may be crucial for the production. Rotors supported by journal bearings are subjected to fluid-induced instability and the occurrence of this phenomenon is influenced by parameters that may vary randomly, so the problem of the identification of the stability threshold is stochastic. This paper applies the Stochastic Collocation method to solve this problem for a given rotor system. The validation of the method is made by the comparison of the results to the results of Monte Carlo simulations for the same problem. The Monte Carlo method requires a great number of simulations for a proper convergence of the results, while the Stochastic Collocation method requires fewer simulations. This difference implies on a considerable processing time difference for the two methods, several hours for the Monte Carlo against some minutes for the Stochastic Collocation. The results of the methods differ on some features: the probability density function generated by the Stochastic Collocation doesn’t fit the normalized histogram generated by the Monte Carlo and the variance of the stability threshold present a considerable difference on the methods. However, the lower and upper limits of the stability threshold on both methods is nearly the same, as well as the mean value for the stability threshold.