1. White paper
Early Detection of Failures in Ultra-Low Temperature (ULT) Freezers
Discovery Analytics for Tomorrow’s Health care
Introduction
One of the key point is to predict the failure(s) of ULT and/or underlying components of
ULT. These ULTs hold biological sample with ages over 30 years – hence making the Reliability
analysis of medical systems or devices as absolutely is an important. ULTs comprise of various
systems and subsystems and are based on the principle of multi-stage refrigeration. A fault or
failure is an inaccurate state of the freezer and the ULTs reliability is a multifarious property that
depends on the functioning of the various parts or components of the ULTs. Data analytics
provide an early detection of a component(s) failure and proper maintenance of ULTs.
Incorrectrefrigerationandstorageofbiologicalsamples,vaccinesandbiomedicalsamples
can lead to loose potency and effectiveness which those become sub-standard can lead to a very
real danger that patients are given a sub-standard vaccine and using sub-standard biological
samples could lead to in accurate insights further research and development of new drug
molecules. The ULTs components need proper diagnostics to maintaining constant temperature.
Therefore sought ways to manage their field failures with early detection and improved
predictability.
Common challenges
The analysis of data from ULTs are non-trivial to incomplete and missing data. Medical device
components fail independent of age, exhibits exponential distribution. Reliability and failure
patterns of a medical device or components may be affected by external factors such as
environmental conditions, operating conditions andother ULTs operating conditions. Reliability
analysis of service data and field data are important or essential to study future occurrence of
ULTs failures.
Data Availability
Failure data information through a multiple sensors is transferred to a database
system and the system stores the field failure data in structured or unstructured level
with incomplete and missing data. Basically data could be in streamline (live data) or a
historical data from a different sources and the data may be retrieved at any time for use
in reliability analysis to figure it out a root cause for failure of ULTs components. Data is
collected from a different log files from a databasesystem and obtainedproductivitydata
and subjected to reliability analysis.
Business Challenge
Data is collected from multiple sensors from ULTs to take preventive measures. Typical
task is, understanding the data and obtaining the correct data prior to statistical analysis.
2. Failures are natural phenomenon; due to internal and external factors like environmental
and operational conditions which resulting the exponential failures. The pharmaceutical
and biomedical companies needed better maintenance and effective storage of samples
at utmost satisfaction of customers. Therefore medical device equipment companies are
very particular about early detection of failures and predict failures in an advance which
improve the reliability of ULTs. We discover hidden patterns in the data (streaming,
historical) using advanced machine learning algorithms. We further correlate the
fault/event occurrence(s) to the hidden patterns, thereby predicting a possible failure
occurrence with a greater confidence.
Approach
Primarily failure data was subjected to descriptive analysis to obtain insights of
failure behavior of ULTs and how environmental and operating conditionsimpact on the
health of the ULTs. Obtained distribution or failure pattern of ULTs for advanced
modeling methods for root cause of the failures. Applied principal approaches for the
reliability of ULTs. Accurate determination of the failure moments were investigated
through various models. Analysis of failure data of a single and multiple systems were
carried out for robust ness of the model for a greater reliability. Statistical tools R, Spark
and Python were used.For predictivemodeling,survivalanalysis modelsand time series
models were applied.
Results
The appropriate statistical modeling methods gave substantial insights to predict
a critical failures in an advance which would help better reliability and proper
maintenance of ULT freezers at utmost satisfaction of customers. Our predictive analytics
are provided better solutions could provide product design in a development of a ULTs.
References
1. Baker R (2001). Data-Based Modeling of the Failure Rate of Repairable Equipment.
Lifetime Data Analysis 7: 65-83.
2. Nelson W (1998). An Application of Graphical Analysis of Repair Data. Quality and
Reliability Engineering International 14(1): 49-52.
3. Robert G. Batson et al (2005). Control Charts for Monitoring Field Failure Data. Quality
and Reliability Engineering International 21: 1-23.