No abstract

No abstract

We consider a general approach to the analysis of the dynamics and errors of different types of micromechanical vibratory gyroscopes, as well as calculation of their performances for application in the design of such gyroscopes. Specifically, we investigate and analyse the dynamics and errors of single-mass gyroscopes, for both translational and rotational movement of the sensitive element. Based on the generalized motion equations, we derive and analyse analytical dependences for basic errors, such as scale factor nonlinearity, bias from misalignment between elastic and measurement axes, and bias from vibrations and dynamic error caused by harmonic angular rate. As a result of dynamics and errors analysis, formulae for the calculation of the main performances are derived, as well as optimal sensitive element design methodologies.

The authors look into the possibility of using artificial neural networks for predicting the deformation characteristics of steels (the parameters of the Basquin-Manson-Coffin strain-life curve equation) based on static strength and plasticity characteristics, by constructing four independent neural networks with different configurations of input and output data. The prediction of parameters of the Basquin-Manson-Coffin equation and the fatigue life calculations by means of artificial neural networks are demonstrated to provide a better accuracy in comparison to the available conventional methods. Keywords: artificial neural networks, parameters of the Basquin-Manson-Coffin strain-life curve equation, fatigue life, static strength and plasticity characteristics.Introduction. In Part 1 [1] we reviewed the currently available methods for assessing the strain-life curves (the parameters of the Basquin-Manson-Coffin equation) and revealed some drawbacks of their definition. Here we will study the possibility of predicting the parameters of the strain-life curves by means of artificial neural networks (ANN).ANN Concept Description. The neurocomputing methodology is a relatively new area of artificial intelligence, where attempts are made to simulate the structure and operation of biological neural systems such as the human brain by constructing ANN in a computer. The use of ANN is especially helpful where one encounters the problems for which the solution is not clearly formulated or where one has to reproduce some mechanism which is otherwise difficult to describe using the available physical models.The methodology of this approach with a comprehensive description of neural networks, considering the multidiscipline nature of this subject, was detailed in [2]. Here we will just outline the ANN background and look into the possibility of using ANN in solving the applied engineering problems.Artificial neural networks consist of simple interconnected elements referred to as the processing elements or artificial neurons, which act as microprocessors. Thus, a neuron represents an information processing unit in a neural network. Figure 1 shows a model of a neuron used as a basis of ANN [2], where three main elements can be identified:1. A set of synapses (or connections), each characterized by its peculiar weight or connection strength. In particular, a signal x j at the input to synapse j connected to neuron k is multiplied by weight w kj . As opposed to brain synapses, the synaptic weight of an artificial neuron can have either positive or negative values.2. The summator adds the input signals weighted relative to the corresponding neuron synapses. This operation can be described as a linear combination.

No abstract