Training problems

Broadband analysis techniques have been used for monitoring the overall mechanical condition of machinery for more than twenty years. The technique is based on the overall vibration or energy from a frequency range of zero to the user-selected maximum frequency, Fmax Broadband data are overall vibration measurements expressed in units such as velocity (PK), acceleration (RMS), etc. This type of data, however, does not provide any indication of the specific frequency components that make up the machine s vibration signature. As a result, specific machine-train problems cannot be isolated and identified. 

Although failure-mode analysis identifies the number and symptoms of machine-train problems, it does not always identify the tme root cause of problems. Root cause must be verified by visual inspection, additional testing, or other techniques such as operating dynamics analysis. 

Table 44.6 is a vibration troubleshooting chart that identifies some of the common failure modes. This table provides general guidelines for interpreting the most common abnormal vibration profiles. These guidelines, however, do not provide positive verification or identification of machine-train problems. Verification requires an understanding of the failure mode and how it appears in the vibration signature. 

Spectrographic analysis allows accurate, rapid measurements of many of the elements present in lubricating oil. These elements are generally classified as wear metals, contaminates, or additives. Some elements can be listed in more than one of these classifications. Standard lubricating oil analysis does not attempt to determine the specific failure modes of developing machine-train problems. Therefore, additional techniques must be used as part of a comprehensive predictive maintenance program. 

Key measurement point locations and orientation to the machine s shaft were selected as part of the database setup to provide the best possible detection of incipient machine-train problems. Deviation from the exact point or orientation will affect the accuracy of acquired data. Therefore, it is important that every measurement throughout the life of the program be acquired at exactly the same point and orientation. In addition, the compressive load or downward force applied to the transducer should also be the same for each measurement. 

All machines have a finite number of failure modes. If you have a thorough understanding of these failure modes and the dynamics of the specific machine, you can learn the vibration analysis techniques that will isolate the specific failure mode or root-cause of each machine-train problem. 

The training problem determines the set of model parameters given above for an observed set of wavelet coefficients. In other words, one first obtains the wavelet coefficients for the time series data that we are interested in and then, the model parameters that best explain the observed data are found by using the maximum likelihood principle. The expectation maximization (EM) approach that jointly estimates the model parameters and the hidden state probabilities is used. This is essentially an upward and downward EM method, which is extended from the Baum-Welch method developed for the chain structure HMM [43, 286]. 

Try to anticipate potential training problems and address them before the training session. An example would be a problem with a direct translation for a term or concept. 

If the employee could do the job if he or she really had to, then a non-training solution will help identify and solve the performance problem. One way to solve the non-training problem is to provide feedback. Feedback should come from... 

Saarineen, S., Bramley R., and Cybenko, G. (1991). The numerical solution of the neural network training problems. CRSD report 1089, Center for Supercomputing Research and Development, University of Illinois, Urbana. 

Sch76] Schmidt, C., Investigation of the Training Problem of Superconducting Magnets, Appl. Phys. Lett., Vol 28, 1976, p. 463-465... 

In the literature [15], a rather flexible trick [12] was introduced. First note that the only way in which the data appeard in the training problem, the formulas (2.26), (2.33) and (2.41), is in the form of dot products, x. x. Now suppose we first mapped the data to some other (possibly infinite dimensional) feature space F, using mapping (see Fig. 2.8) ... 

The NanoCell training problem as stated is extremely difficult. We are just now beginning to make progress on this problem using neuro-dynamic programming. These initial results are encouraging, but before attacking this... 

The simplified NanoCell training problem is particularly well suited to genetic algorithms. After presenting the fundamentals of genetic algorithms, a heuristic for solving this optimization problem is presented. 

In adapting the GA to the NanoCell training problem, the fitness function is the most difficult issue. First consider the voltage in - current out setup. Recall that Iql and Iqh denote the high and low output current thresholds. 

---