The document discusses a project to examine factors that lead to long-term performance deterioration of sensor systems and assess the reliability of sensor components versus total systems. It aims to explore opportunities for DNV GL in sensor system reliability, including verification of sensor performance and assessment of risk. Key factors that affect sensor reliability are discussed, as well as methods to improve reliability through design, maintenance, and reliability evaluation.

1. DNV GL © 2013
Ungraded
25 November 2016 SAFER, SMARTER, GREENERDNV GL © 2013 AM002
Ungraded
25 November 2016
Reliable Sensor Technologies (REST)
1

2. DNV GL © 2013
Ungraded
25 November 2016
Scope of REST Project
 To examine factors that lead to long-term performance deterioration of sensor
systems.
 To assess the reliability of the sensor components vs total system (power,
communications, data processor, and inter-connections).
 To investigate using modelling analysis (e.g. Bayesian networks) for predicting
and evaluating sensor system reliability.
 To explore business opportunity for DNV GL in the field of sensor systems
reliability.
2

3. DNV GL © 2013
Ungraded
25 November 2016
Importance of REST
3
 Digitalization
– Sensor Technology itself is an important part of Digitalization.
– DNV GL relies on reliable sensor data for surveying, inspection, monitoring, digital approval,
risk assessment etc.
 Business Opportunities
– Verification/validation of sensors performance; Prediction of sensor reliability; Assessment of
risk.

4. DNV GL © 2013
Ungraded
25 November 2016
Definition
 A sensor is a device that measures a physical
quantity by generating a functionally related
output which can be read by an observer or
by an electronic instrument.
 A working sensor system is a composite of
four distinct parts:
– Sensing element produces electrical response
when stimulated by external factors;
– Signal conditioning element modifies and
processes the electrical signal to be understood
properly by the receiver;
– Sensor interface allows the device to acquire,
store, and communicate with an external
interface;
– Power system provides external power resources
or to harvest energy.
4

5. DNV GL © 2013
Ungraded
25 November 2016
Sensor System Reliability
 The ability of a sensor system to perform its
required functions under stated conditions for a
specified period of time. For this reason, the
reliability of a sensor system will be strongly
dependent upon its age, context and application.
 A high reliability is especially critical for the
following application scenarios:
– Complex and smart sensor systems (e.g. Sensor Fusion)
– Highly integrated miniature sensor systems (e.g. MEMS
sensors)
– Long term monitoring requirements (e.g. Condition
Monitoring)
– High Consequence/Mission Critical applications (e.g.
Leak Detection)
– Application requires dynamic response (e.g. DPS)
– Harsh and extreme working environments (e.g. HTHP).
5

6. DNV GL © 2013
Ungraded
25 November 2016
Factors Affecting Sensor Reliability
 Each component of the sensor system, design, manufacturing, packaging,
installation, maintenance, calibration, and the software used by a sensor system
can affect its reliability.
 One method for assessing sensor system reliability is Failure Mode, Effects and
Criticality Analysis (FMECA), where a risk assessment is made of each failure
mode to determine its criticality. Criticality is derived from an assessment of the
probability that a particular failure will occur combined with the severity of the
failure if it does occur (i.e. the consequence).
6

7. DNV GL © 2013
Ungraded
25 November 2016
Generic Ranking of a Pressure Sensor
 To carry out an FMECA for a sensor system, a systematic analysis need to be
performed based on the sensing mechanism, the working environments and the
overall redundancy.
 Table below shows the criticality of a pressure sensor (generic) used in subsea
processing, where the criticalities were divided into five categories of Very Low,
Low, Medium, High, and Very High.
7
Sensor System Components Failure Modes Criticality
Very low Low Medium High Very High
Sensor elements Degradation of Sensing Material
Thermal Induced
Signal Processing
Components
Degradation of contacts
or connections
Thermal Induced
Degradation of Signal Processors
Power System
Degradation of contacts
or connections
Faulty Electronics
Loss of Power
Interface Communicating
Components
Degradation of contacts
or connections
Faulty Electronics
Poor/Insecure Signal

8. DNV GL © 2013
Ungraded
25 November 2016
Improving Sensor System Reliability
 Design/Manufacturing.
 Choosing the Suitable Sensor Systems.
 A Strictly Enforced Maintenance and Calibration Plan.
8

9. DNV GL © 2013
Ungraded
25 November 2016
Design/Manufacturing
 The most effective strategy to improve sensor system reliability is through careful
system design and well-controlled manufacturing quality.
 A current tendency is to rely on a single vendor who can perform the entire
process of sensor design, manufacturing, and testing.
 Example: A variable capacitance silicon accelerometer used in implantable devices
with no field failures among the four million parts was designed and manufactured
using MEMS technology.
9

10. DNV GL © 2013
Ungraded
25 November 2016
Choosing the Suitable Sensor Systems
 The reliability of a sensor system will be strongly influenced by the working
environment into which it is applied.
 For mission critical purposes or long term usage such as condition monitoring of
infrastructures, certain levels of testing (normally including destructive testing)
are necessary before deploying sensors into the field.
 There have been attempts to develop computer based models such as “Multi
Criteria Decision Making” models for aiding sensor selection.
 In practice, the majority of sensor selections is based on manufacturer provided
technical data sheets, sensor users’ past experience, or expert knowledge and
recommendations.
10

11. DNV GL © 2013
Ungraded
25 November 2016
Choosing the Suitable Sensor Systems
Classification Societies Roles:
 DNV GL’s Recommended Practices on
Reliable Sensor System for Maritime
Monitoring.
 Objective: to lead to more reliable sensor
systems and subsequently improved quality of
the data collected for analysis and decision
support in condition and performance
monitoring of equipment, environmental
monitoring and sensing systems
Methodology: to ensuring system level
reliability through defining the minimum
requirements of the lowermost level, i.e.
sensor components in a future maritime
condition monitoring system.
11
Component
Failure modes
Sensors
Output
Component Sensor Requirements
(Reliable Sensor Technology

12. DNV GL © 2013
Ungraded
25 November 2016
A Strictly Enforced Maintenance and Calibration Plan
 A practical maintenance plan must be determined based on previous experience
and intensive testing in both lab and field to ensure that sensor systems can be
used with high reliability, efforts must be taken to eliminate human errors.
 After deployment, sensors require occasional testing and replacement of wear-out
components. Most sensor manufacturers will provide guidance for the
maintenance requirements, but users have to adopt these requirements into their
own procedures.
 Calibration is essential for establishing the accuracy of a sensor in relation to
standards. Some modern “smart” sensing systems have the capability of self-
calibration due to integration with an ASIC (Application-Specific Integrated
Circuit).
12

13. DNV GL © 2013
Ungraded
25 November 2016
Evaluation of Sensor System Reliability
 Testing can be performed at different levels depends on the application scenarios.
 Testing is to verify the reliability of a sensor system, but can not improve it.
 Accelerated Life Testing (ALT) is typically employed to assess the long-term
performance of a sensor in a short period of time.
 ALT stresses the sensor systems by exposure to purposely harsh environments to
induce field failure at a much faster rate, which should include:
 1. Define the scope
 2. Collect required information
 3. Determine the mechanisms of degradation
to establish appropriate types and levels of stresses
 4. Conduct the accelerated tests
 5. Predict the field life of the sensor system
13

14. DNV GL © 2013
Ungraded
25 November 2016
Prediction of Sensor System Reliability
 Four categories of detection methods can applied to
develop algorithms to detect faulty sensor data as a
screening tool prior to passing information on to
decision-aid tools:
– Rule-based methods define heuristic rules/constraints
that the sensor readings must satisfy.
– Estimation methods define “normal” sensor
behavior by leveraging spatial correlation in
measurements at different sensors.
– Time series analysis based methods compare a sensor
measurement against its predicted value based on time
series forecasting to determine if it is faulty .
– Learning-based methods infer statistically established
models to identify faulty sensor readings using training
data.
14

15. DNV GL © 2013
Ungraded
25 November 2016
Prediction of Sensor System Reliability
 There exists the need to estimate the
uncertainty attached to the sensor data being
received, useful in the situation where sensor
systems can degrade over time in service,
and thus the data may need to be corrected
or compensated before decisions can be
made.
 Knowing enough information about a sensor
system, including the working mechanism of
sensors, its technical specifications, failure
modes and working environments, it is
feasible to construct a Bayesian network
model for filtering and assessing unreliable
sensor data.
1
Generic structure of a simplified Bayesian network to predict the reliability of sensor
data under the influence of degradation and aging of the sensor components.

16. DNV GL © 2013
Ungraded
25 November 2016
A Case Study: Digital Downhole
 Digital downhole refers to the entire instrumented drill string system, the sensor
communication system, and the digital twin model.
 Sensor systems assessing the conditions downhole and reporting them to the
operations managers.
 Sensors of significance include pressure, flow, temperature and fluid chemistry
monitoring sensors.
16
The sensor data flowing back to the
operations center provides the inputs
needed to construct the real-time
digital representation (i.e. the digital
twin) of the downhole system, so that
operators see an exact picture of the
state of health, and can make
adjustments to correct any deviations
or initiate emergency procedures.
The Digital Downhole instrumented by pressure and flow sensors

17. DNV GL © 2013
Ungraded
25 November 2016
A Case Study: Digital Downhole
 System Constraints on Sensor Data
 Physics engine places constraints upon “normal sensor data” based on type and
position:
 Internalizing these constraints allows cross-checking via analytical redundancy
relations
17
Downhole pressure (pd) linked to
pump pressure (pp), hydraulic
pressure (gravity, Gd) and friction
in drillstring (friction factor θd and
flow, q)
Annular pressure (pa) linked to
downhole pressure (pd), and
friction across the bore (friction
factor θb and flow, q)

18. DNV GL © 2013
Ungraded
25 November 2016
 Fault isolation and detection in digital
downhole.
 Using Analytical Redundancy Relations
to generate digital “fingerprints” for
normal operations versus failure
modes.
 Each failure mode for the downhole
system will correspond to a unique
pattern of deviation in analytical
redundancy relations. These unique
patterns can be used as fingerprints
that allow fault detection and isolation
within the Digital Downhole system.
 Data smoothing and p-value hypothesis
testing required to discern true
constraint-breaking from noise.
A Case Study: Digital Downhole
18
Sensor issues Process issues

19. DNV GL © 2013
Ungraded
25 November 2016
A Case Study: Digital Downhole
19
 Combined survival probability models
with fingerprinting and the Digital Twins
technology for sensor reliability
assessment.
 As the Digital Downhole system collects
data and reports on sensor
performance through the use of
analytical redundancy relations, the
sensor reliability models can be
updated.
 Conversely, the updatable sensor
reliability models can be used to assess
the likelihood of a sensor failure. (i.e.
to distinguish a true constraint-
breaking event from the background
noise.)
Sensors reliability model constructed through
lab-based testing, but are then updated in real-
time by data coming in through “fingerprint”
analysis performed by the Digital Twin

20. DNV GL © 2013
Ungraded
25 November 2016
A Potential Business Case
20
• With digital communication:
-Either there is communication or not.
There is no accuracy reduction along
the signal chain.
- Only degradation is inside the sensor.
• Pressure Sensor:
+/- 0.35 Bar when “out of the box”
+/- 9.1 Bar after 25 years
• Temperature Sensor:
+/- 0.6 C “out of the box”
+/- 5.6 C after 25 years
WEPS-100
Series Subsea
Sensors Pressure
and Temperature
by SIEMENS
• The mechanism leads to the huge drift is unclear
but may be related to operation temperature etc.
• An unplanned maintenance of
each module could cost Akers millions $ per day.

21. DNV GL © 2013
Ungraded
25 November 2016
A Potential Business Case
DNV GL can help through:
 Consulting with sensor manufacturer on product details and test results.
 Consulting with end-users on operation condition and field test results.
 Creating a Bayesian networks model integrating all factors resulting in sensor drift.
 Predicting the effect of sensor drift in field operation, i.e. what is the “real” reading.
 Testing to validate and improve the models.
21
Year

22. DNV GL © 2013
Ungraded
25 November 2016
Potential Business Models: Verification and Classification
 Control/Mitigate Risk for Customer:
- Accurate interpretation of the sensor data (considering the sensor reliability);
- Better selection of sensors (predicting the sensor reliability in the field);
- Modify the sensor life curve by considering the sensor degradation mechanism and
operation conditions etc.
- Create models from top down, i.e. from inputs to results.
 Verify/ Qualify Sensors for certain applications:
– Using customer provided sensor data as input to an BN model for validation.
– Output from the model: * Whether sensor accuracy is within range of actual value and its
frequency. * Whether the signal is a false positive and its frequency, and *Whether the
signal quality or accuracy degrades with time.
 Explore methods for sensor system reliability assessment as a base for future
certification services.
– Type approval of sensors for condition monitoring purpose, where sensor data transferred
to class is part of a (semi)real-time survey scheme: CMC.
– Outline Class notation on newbuilds and retrofits
22

23. DNV GL © 2013
Ungraded
25 November 2016
SAFER, SMARTER, GREENER
www.dnvgl.com
23
Shan Guan, Shan.Guan@Dnvgl.com