skip to content

Data Analysis Technology

The key to successful condition monitoring is the analysis of the data. The core of Ventech Systems' technology lies in its data analysis capability, and in particular its unique approach to the automated screening of raw data.

The Problem of Complex Data

Vibration measurements taken from rotating machinery provide a very rich source of information regarding the health of the system. By appropriate processing of the data, virtually all of the major drive train defect types can be identified, often at a very early stage of development. The basic technology has been proven over many decades, and has come to be regarded as an absolutely essential part of plant operation and maintenance in many different industrial sectors.

Whilst the richness of this type of data is what makes it so useful, it also presents its own problems. By its very nature, this type of data tends to be complex, and the datasets can be extremely large. Because of this, the useful information within the data can often be difficult to locate, and the critical signatures can easily be masked by less interesting features. Considerable expertise is usually required to process the data, such that the useful and interesting information can be successfully extracted.

When applied to the remote monitoring of wind turbines these problems become particularly acute. More than any other factor, this is due to the extremely variable operational envelope that turbines experience in service. This introduces a lot of additional variability into the data when compared with the majority of applications, further complicating the analysis. The fact that the gearbox is invariably a multistage unit where the critical components are often not directly accessible, which very commonly employ planetary stages and is mounted on a (relatively) flexible bedplate creates even more difficulties. The reality is that wind turbines present one of the most challenging industrial applications for this type of monitoring.

Data Screening

To cope with the complexity and the sheer quantity of data generated in vibration based condition monitoring applications, virtually all systems employ some form of automatic 'screening' of the data. This involves applying one or more algorithms to the raw data, the purpose of which is to provide an initial (automated) indication of where defects may be present. In the simplest case, these algorithms take the form of checking a particular feature (the amplitude of a specific frequency component is a common example) against a predefined threshold. If the algorithms indicate that a defect may be present, then the data will be subject to a more detailed manual investigation; otherwise they will not be examined further. Since in the great majority of cases the machinery will be perfectly healthy, the quantity of data to be analysed is reduced dramatically, thus bringing the workload for the human analyst down to manageable proportions.

Clearly, the quality of the output provided to the operator depends critically on how effective these algorithms are at correctly identifying fault indications within the data. If the algorithms are too sensitive, then large quantities of false indications will be produced, swamping the analyst and making the process unmanageable. If the algorithms are not sensitive enough, then faults may be missed, in the worst case leading to catastrophic failure. The reliability and accuracy of these algorithms is perhaps the single most important factor in determining the success or otherwise of a condition monitoring programme.

Ventech Systems Data Analysis Technology

Drawing on many years of research supporting the development of condition monitoring technologies in the aerospace sector, Ventech Systems have developed a unique approach to the screening problem. This simultaneously reduces the false positives and false negatives to levels not previously achievable, thus making the screening process much more reliable. By supplying more reliable indications to the operator earlier in the degradation process, we can maximise the opportunity for cost effective management of the situation.

The approach involves a two-step process. The first step utilises a series of individual, highly sophisticated screening algorithms which are designed to detect very specific types of indications. Much of this work draws heavily on our experience in the aerospace sector, where this type of algorithm has been in development for decades. The second step involves a data fusion approach, where we consider the results from the first stage both as individual indicators and in relation to each other. It is this second step which provides the very high degree of fault detection reliability characteristic of the overall approach.

back to top