Ondřej Pokora

Contribution to system failure occurrence prediction and to system remaining useful life estimation based on oil field data

At present, numerous approaches are devoted to monitoring a system state. Their intention is to determine the current state of a system and predict reliability parameters for the future. This article addresses one of the several possible approaches that allows us to determine a system technical state on the basis of diagnostic data. These diagnostic data are from the area of tribiodagnostics, namely, engine oil. The article examines iron and lead particles that are selected deliberately with respect to their origin in kinematic parts of the system and their degree of correlation with operation measures. The particles occur in oil during both operating time and calendar time development. To model their occurrence during operation time, we have used, in the first part of the article, a mathematical regression method to set parameters. In the second part, we have applied a diffusion model based on a Wiener process. The results confirm that we are able to estimate the residual technical life of a system. Moreover, the results enable us to schedule properly the intervals of preventive maintenance (oil change) and to plan a mission/operation. This results in optimising life cycle costs. It is assumed that the potential of the diagnostic data will be extracted by other approaches and methods. In the subsequent work, it will be useful to determine specific interval values of optimised preventive maintenance.

Perspective analysis outcomes of selected tribodiagnostic data used as input for condition based maintenance

Abstract The aim of this article is to estimate system soft failure occurrence and residual technical life in order to optimise firmly planned preventive maintenance. To do this, selected wear particles from oil field data are analysed. Using a large tribodiagnostic dataset we estimate the residual technical life of the observed system statistically. The dataset includes information about particles contained in oil which testify to oil conditions as well as system conditions. We focus here on the wear particles which we (and other analysts) consider to be interesting, ferrum (Fe) and lead (Pb), regarded as contact degradation and wear products. By modelling the occurrence of particles in oil we plan to determine the expected moment when soft failure occurs; this moment might then be set as the time to perform preventive maintenance (PM). Both operation time and calendar time are used here for modelling, for soft failure occurrence determination and for residual technical life estimation. The modelling is based on the specific characteristics of two diffusion processes, the Wiener process (WP) with positive drift and the Ornstein–Uhlenbeck process (OUP). We also applied a fuzzy inference system to support our first results from the diffusion processes as there is a level of uncertainty and fuzziness in the oil field data. Following the modelling outcomes we are able to judge the system hazard rate, predict expected mean residual life and set up principles of “condition based maintenance” (CBM) for this system. However, the possible uses of our outcomes are much wider. For example, they can be used as inputs for operation and mission planning, and life cycle costs can be significantly reduced thanks to the maintenance optimisation.

Statistical approach in search for optimal signal in simple olfactory neuronal models

Several models (concentration detectors and a flux detector) for coding of odor intensity in olfactory sensory neurons are investigated. Behavior of the system is described by different stochastic processes of binding the odorant molecules to the receptors and their activation. Characteristics how well the odorant concentration can be estimated from the knowledge of response, the number of activated neurons, are studied. The approach is based on the Fisher information and analogous measures. These measures of optimality are computed and applied to locate the odorant concentration which is most suitable for coding. The results are compared with the classical deterministic approach which judges the optimal odorant concentration via steepness of the input-output function.

Stimulus-response curves in sensory neurons: how to find the stimulus measurable with the highest precision

To study sensory neurons, the neuron response is plotted versus stimulus level. The aim of the present contribution is to determine how well two different levels of the incoming stimulation can be distinguished on the basis of their evoked responses. Two generic models of response function are presented and studied under the influence of noise. We show in these noisy cases that the most suitable signal, from the point of view of its identification, is not unique. To obtain the best identification we propose to use measures based on Fisher information. For these measures, we show that the most identifiable signal may differ from that derived when the noise is neglected.

System Condition Estimation Based on Selected Tribodiagnostic Data

The aim of the paper is to estimate a system-soft failure occurrence and residual technical life. When estimating a residual technical life statistically, usually a big amount of tribodiagnostic data is used. Data include the information about particles contained in oil that testifies to oil and system conditions. We focus here on the particles that we consider to be interesting. They are ferrum (Fe) and lead (Pb) as contact degradation product. By modelling the occurrence of particles in oil, we expect to determine the expected moment for soft failure occurrence or adequate moment to perform preventive maintenance. The way of our modelling is based on the specific characteristics of diffusion processes, namely the Wiener process with positive drift and Ornstein–Uhlenbeck process. Following the modelling results, we could judge hazard rate and set-up principles of ‘CBM - Condition Based Maintenance’ (CBM). However, the possibilities are much wider, because we can also plan operation, mission and reduce life cost. Copyright

Ergodicity and parameter estimates in auditory neural circuits

This paper discusses ergodic properties and circular statistical characteristics in neuronal spike trains. Ergodicity means that the average taken over a long time period and over smaller population should equal the average in less time and larger population. The objectives are to show simple examples of design and validation of a neuronal model, where the ergodicity assumption helps find correspondence between variables and parameters. The methods used are analytical and numerical computations, numerical models of phenomenological spiking neurons and neuronal circuits. Results obtained using these methods are the following. They are: a formula to calculate vector strength of neural spike timing dependent on the spike train parameters, description of parameters of spike train variability and model of output spiking density based on assumption of the computation realized by sound localization neural circuit. Theoretical results are illustrated by references to experimental data. Examples of neurons where spike trains have and do not have the ergodic property are then discussed.

Estimation of residual life based on vehicle tribo data

The aim of the article is to estimate a system technical life. When estimating a residual technical life statistically, a big amount of tribo-diagnostic data is used. This data serves as the initial source of information. It includes the information about particles contained in oil which testify to oil condition as well as system condition. We focus on the particles which we consider to be interesting. This kind of information has good technical and analytical potential which has not been explored well yet. By modelling the occurrence of particles in oil we expect to find out when a more adequate moment for performing preventive maintenance might come. The way of modelling is based on the specific characteristics of diffusion processes, namely the Wiener process. Following the modelling results we could in fact set the principles of “CBM - Condition Based Maintenance”. However, the possibilities are much wider, since we can also plan operation and mission. All these steps result in inevitable cost saving.

Steady-state properties of coding of odor intensity in olfactory sensory neurons

Several models for coding of odor intensity in olfactory sensory neurons are investigated. Behavior of the systems is described by stochastic processes of binding (and activation). Characteristics how well the odorant concentration can be estimated from the knowledge of response, the concentration of bounded (activated) neuron receptors, are studied. This approach is based on the Fisher information and analogous measures. These measures are computed and applied to locate the coding range, levels of the odorant concentration which are most suitable for estimation. Results are compared with the classical (deterministic) approach to determine the coding range via steepness of the input-output transfer function.