Identifying Failure Modes

IDENTIFYING FAILURE MODES
RETROSPECTIVELY USING RCM DATA

J. Sikorska (Imes Australia Pty Ltd),
L. Hammond (DSTO),
P. Kelly (CASWA Pty Ltd)

SUMMARY: Managers are constantly being told that the quality of data is insufficient for producing
reliable results from business models. Reliability analysis is no exception. In this paper, the authors
show that it is possible to retrospectivelyimprove the quality of failure histories stored in computerised
maintenance management systems (CMMS). A tool was developed to help reliability professionals
assign the most appropriate failure mode to a particular work order. Case-based reasoning is used to
contextualise the data and reduce the number of possible options to a list of most likely candidates; the
case-bank is developed from FMEAbased studies such as RCM. Traditionally, CMMS and FMEA
systems have rarely been integrated and differences in emphasis, underlying coding structures and
information supplied often make this difficult. However, using diesel engine data from HMAS Anzac, it
is shown that these issues can be overcome in a sustainable manner. It is also possible to adapt the system
for use in real-time failuretroubleshooting.
KEYWORDS: FMEA, RCM, CMMS, failure modes, reliability analysis, data quality, data cleansing.
1.

INTRODUCTION

It is widely accepted that data is fundamental to robust and reliable decision-making. In a previous paper [1] the authors
developed an eight step process for improving data quality. This process is based on the basic definition of data quality as:
‘Data that is fitfor purpose’. The paper suggested that a DQ process should: (1) define the business need; (2) identify metrics;
(3) determine the data required; (4) analyse quality (data profiling/auditing); (5) identify ways to improve quality; (6)
implement changes (data cleansing); (7) assess changes; and finally (8) establish review periods. All but steps 4, 5 and 6
should be achievable using anorganisation’s standard business processes. Due to their very specific application, advanced
systems and processes for data profiling and data cleansing however, are not commonly available.
Wherever possible, the best and easiest option to rectify poor data quality is to monitor, report and prohibit discrepancies as the
data is being entered, thus allowing or compelling data collectors to rectify theerroneous information immediately (i.e. before
they forget and/or the data is no longer available). There is no doubt that this should be a mandatory part of any data capture
process. There is also merit in developing mechanisms to assess how well software manages the data entry process. However,
it is unrealistic to expect that any datasets will ever be completely error free, no matter how wellintentioned the data collectors
and well managed the associated systems, processes and databases [2]. This approach also disregards the wealth of
information that presently resides in existing data stores. Therefore, when data integrity is crucial, subsequent profiling and
improvement techniques are also required.
This paper describes a method for cleansing data retrospectively. Although thiswork relates to failure information being used
for reliability analysis, this approach may be applied to other types of information.

2.

2.1

BACKGROUND

Problem Statement

A business need to undertake CBM and reliability modelling was identified. Associated metrics require historical failure data,
which is ostensibly stored in a corporate computerized maintenance management system(CMMS). Minimum data for basic
reliability/CBM analysis is:
1. What component failed?
2. What was the component’s failure mode?
3. What system was affected?
4. How was it affected?
5. What was the component’s age at failure?
If a system is appropriately documented (e.g. by using reliability block diagrams), then by knowing the specific component
failure mode it is possible to deduce...