Reliability Study on a Fire Detection

System

Sebastian Braendle
29/06/2020 EDMS: 2390547 v.1
Agenda

1. Calculation Model
2. Reliability Data
3. Calculation Framework
4. Calculation Example
5. Opportunities and Limitations

29/06/2020 2
Calculation Model
Calculation Model – Most Significant Detection System

Requirements for the considered fire detection system:

• Well known and documented
• Commonly used at CERN and reasonable for the future
→ To cover as much detection systems as possible with the study

Chosen system:
• Securifire (Securiton), installed since 2017
• Workshop for the production of circuit boards (CERN bldg. 107)
• High complexity: up to 216 components distributed over 4 floors
• Contains several types of smoke detectors

29/06/2020 4
Calculation Model – Assumptions

Assumptions
• The model resolution ends up at the component level (for e.g. a smoke detector is considered as a closed part)
• If more than one component of a type occurs in a system they are considered as redundant
• If one component type fails, the whole system fails
• Component failures occur random and independent (failures exp. distributed)
• In case of a failure, the system will be repaired within X hours (repairs exp. distributed)

System: Room 1-403 of Bldg. 107

Component types:
• Control panel: 1
• Optical point detectors: 2
• Alarm siren: 1

29/06/2020 5
Calculation Model – Component Identification

Possible component types:

• Control panel
• Optical point detector (smoke detector)
• Aspirating detector (smoke detector)
• Venturi detector (duct smoke detector)
• Alarm siren

Characteristics:
→ System layout and arrangement of component
types stays for different rooms/ buildings the same
→ Calculation model increases/ decreases with the
amount of components of each component type

29/06/2020 6
 Calculation Model - Generic Fault Tree Layout
Legend: Fault Tree
CP Control panel
OPD Optical point detector FSystem Failure probability of
ASD Aspirating smoke detector the whole system
VSD Venturi smoke detector
AS Alarm siren OR
PC Problem code (failure rate)

Failure probability of
CPType OPDType ASDType VSDType ASType every component type

AND AND AND AND

∙∙∙
Failure probability of a
CP1 OPD1 ∙∙∙ OPDn ASD1 ∙∙∙ ASDn VSD1 ∙∙∙ VSDn AS1 ASn
single component

OR OR OR OR OR

Failure probability of a
PCCP_1 ∙∙∙ PCCP_n PCOPD_1 ∙∙∙ PCOPD_n PCASD_1 ∙∙∙ PCASD_n PCVSD_1 ∙∙∙ PCVSD_n PCAS_1 ∙∙∙ PCAS_n specific failure on a
single component

Legend: Calculation
F(t) Failure probability 1
λ Failure rate 𝐹(𝑡)𝑆𝑦𝑠𝑡𝑒𝑚 = 1 − 1 − 𝐹𝐶𝑃𝑇𝑦𝑝𝑒 1 − 𝐹𝑂𝑃𝐷𝑇𝑦𝑝𝑒 1 − 𝐹𝐴𝑆𝐷𝑇𝑦𝑝𝑒 1 − 𝐹𝑉𝑆𝐷𝑇𝑦𝑝𝑒 1 − 𝐹𝐴𝑆𝑇𝑦𝑝𝑒 = 1 − 𝑒 −𝜆𝑆𝑦𝑠𝑡𝑒𝑚∙𝑡 𝐸𝑋 = = 𝑀𝑇𝐵𝐹
𝜆𝑆𝑦𝑠𝑡𝑒𝑚
MTBF Mean time between failure

29/06/2020 7
Calculation Model – Availability of the System
Legend: Calculation
MTBF Mean time between failure
Availability of the system (steady state): MTTF Mean time to repair
AD Availability (steady state)

System status 𝑀𝑇𝐵𝐹

𝐴𝐷 = lim 𝐴(𝑡) =
𝑡→∞ 𝑀𝑇𝐵𝐹 + 𝑀𝑇𝑇𝑅 Assumptions:
MTBF MTTR
• Failures occur exponential distributed with the
constant failure rate λ
Operational • Repairs occur exponential distributed with the
constant repair rate μ

Under repair
Time

→ The availability is the probability that a system is in functional condition

29/06/2020 8
Reliability Data
Definition of a Failure

Failure of components:
• If the component itself has a malfunction (for e.g. “detector fault”)
• If the component is not able to fulfill its foreseen function and need a repair

Meaning for a system failure:

• The failure probability of the whole system is a combination of:
• Components
• Infrastructure (cable for power supply and communication)
• Layout (redundant components)
• Environmental and operational conditions

Approach: Using pre defined closing codes/ problem codes to track failures in Infor

29/06/2020 10
 Problem Codes – Which codes are relevant? Failure
No failure

Control Panel Optical Point Detector Aspirating Smoke Detector Venturi Detector Alarm Siren
Problem SF-SECURIFIRE SF-MCD573X SF-ASD535 SF-LKM593X S-SIR745/F
Description (fr) Description (en)
Code tOperational: 385104 h tOperational: 18943560 h tOperational: 8289552 h tOperational: 55440 h tOperational: 7961160 h
Amount: 36 Amount: 2129 Amount: 729 Amount: 5 Amount: 671
Défaut d'isolement Insulation fault SP001 2
Défaut batteries Battery fault SP002 2
Défaut chargeur de batteries Battery charger fault SP003 1
Coupure alimentation électrique Power supply cut-off SP004 7
Défaut composant centrale (Carte, BG, Sirene …) Central component fault (Card, BG, Sirene, Flash...) SP005 3
Déclenchement organe de protection Protective device release SP006
Alarme sur détection réelle motivée par Alarm on actual detection motivated by
SP007 1 1
Fumée/Chaleur/Evac/Gaz/Explosion Smoke/Heat/Evac/Gas/Explosion
Défaut détecteur Detector fault SP008 1 2 1 1
Défaut pompe Pump fault SP009
Prealarme gaz determinée (sauf détection réelle) Determined gas pre-alarm (except real detection) SP010
Alarme gaz determinée (sauf détection réelle) Determined gas alarm (except real detection) SP011
Alarme feu determinée (sauf détection réelle) Determined fire alarm (except real detection) SP012 1 2 1
Alarme evac determinée (sauf détection réelle) Determined evacuation alarm (except real detection) SP013
Defaut surveillance debit aspiration Suction flow monitoring fault SP014 3 64
Defaut ligne ou boucle de detection Line fault or detection loop SP015 5 2 2
Suite intervention non finalisee astreinte Continued intervention not finalized on-call SP016
Autres Others SP017 7 4
Prealarme gaz non determinée Gaz prealarm not determined SP018
Alarme gaz non determinée Undetermined gas alarm SP019
Alarme feu non determinée Undetermined fire alarm SP020 2
Alarme evac non determinée Evac alarm not determined SP021
EDMS: 1706394 “CODIFICATION GMAO DES MATERIELS POUR LES SYSTEMES D’ALARMES“ Database query time span: 1/01/2019 – 10/05/2020

29/06/2020 11
 Calculation Model - Generic Fault Tree Layout
Legend:
CP ... Control Panel
OPD ... Optical Point Detector
ASD ... Aspirating Smoke Detector
VSD ... Venturi Smoke Detector
AS ... Alarm Siren
PC ... Problem code FSystem Failure probability of
the whole system

OR

Failure probability of
CPType OPDType ASDType VSDType ASType every component type

AND AND AND AND

∙∙∙
Failure probability of a
CP1 OPD1 ∙∙∙ OPDn ASD1 ∙∙∙ ASDn VSD1 ∙∙∙ VSDn AS1 ASn
single component

OR OR OR OR OR

Failure probability of a
SP001 SP002 SP015 SP001 SP004 SP008 SP015 SP001 SP003 SP008 SP015 SP001 SP004 SP008 SP015 SP001 SP004 SP015 specific failure on a
SP003
SP004 single component
SP002 SP004 SP009

29/06/2020 12
 Operational Data Evaluation
Operational Data Reliability Parameter Estimation Quantifying the Uncertainty

Database query Weibull analysis Confidence interval

2 (Chi-square) distribution

Operational time ti
Amount of failures #

29/06/2020 13
Calculation Framework Overview
 Calculation Framework – Overview
Input Calculation Output
Availability calculation:

System layout properties:

System availability:
• Component types
• Quantity of components
• Mean time to repair MTTR 𝐴𝑆𝑦𝑠𝑡𝑒𝑚
CP: 1
OPD: 2
AS: 1

Fault tree model:

Operational data: Failure rate estimation: • Calculation
• Database query • Weibull analysis

29/06/2020 15
 Example – Room 1-403 of building 107
FSystem

System properties: OR

Component type Amount

Control panel 1 CPType OPDType ASType
Optical point detector 2
AND
Aspirating smoke detector 0
Venturi smoke detector 0
Alarm siren 1 CP1 OPD1 OPD2 AS1
Assumed downtime due to an
intervention: 2h OR OR OR OR

SP002
Expected mean time between failure: SP001 SP003 SP015 SP001 SP004 SP008 SP015 SP001 SP004 SP008 SP015 SP001 SP004 SP015
SP004
1 ℎ
𝑀𝑇𝐵𝐹 = = 54962
𝜆𝑆𝑦𝑠𝑡𝑒𝑚 𝑓𝑎𝑖𝑙𝑢𝑟𝑒

Expected mean time to repair:

ℎ 𝐹(𝑡)𝑆𝑦𝑠𝑡𝑒𝑚 = 1 − 1 − 𝐹𝐶𝑃𝑇𝑦𝑝𝑒 1 − 𝐹𝑂𝑃𝐷𝑇𝑦𝑝𝑒 1 − 𝐹𝐴𝑆𝑇𝑦𝑝𝑒 = 1 − 𝑒 −𝜆𝑆𝑦𝑠𝑡𝑒𝑚 ∙𝑡
𝑀𝑇𝑇𝑅 = 2
𝑓𝑎𝑖𝑙𝑢𝑟𝑒

29/06/2020 16
 Example – Room 1-403 of building 107

Availability of the system:

System status

ℎ ℎ
𝑀𝑇𝐵𝐹 = 54962 𝑀𝑇𝑇𝑅 = 2
𝑓𝑎𝑖𝑙𝑢𝑟𝑒 𝑓𝑎𝑖𝑙𝑢𝑟𝑒
ℎ
𝑀𝑇𝐵𝐹 54962
𝑓𝑎𝑖𝑙𝑢𝑟𝑒
𝐴𝐷 = = = 99.9964% Operational
𝑀𝑇𝐵𝐹 + 𝑀𝑇𝑇𝑅 54962 ℎ ℎ
+2
𝑓𝑎𝑖𝑙𝑢𝑟𝑒 𝑓𝑎𝑖𝑙𝑢𝑟𝑒

Under repair
Time

→ There is a 99.9964% probability that the considered fire detection

system will be available

29/06/2020 17
Influencing Factors
Model: Example 1 Example 2
Control panel 1 1
Optical point detector 2 -
Aspirating smoke detector - 1
Venturi smoke detector - 1
Alarm Siren 1 1
Mean time between failure [h] 54962.85 27327.53
Mean time to repair [h] 2 20 100 2 20 100
System availability [%] 99.9963 99.9636 99.8184 99.9927 99.9269 99.6354

𝑀𝑇𝐵𝐹 1
𝐴𝐷 = =
𝑀𝑇𝐵𝐹 + 𝑀𝑇𝑇𝑅 1 + 𝑀𝑇𝑇𝑅 MTTR ↑ ... AD↓
𝑀𝑇𝐵𝐹 MTBF ↑ ... AD↑

29/06/2020 18
 Detector (un)availability due to IS37 disabling in 2019

𝑑𝑖𝑠𝑎𝑏𝑙𝑒𝑑 𝑑𝑒𝑡𝑒𝑐𝑡𝑜𝑟 𝑡𝑖𝑚𝑒 σ𝑛𝑖=1 𝑡𝑖𝑆𝑇37 25164 𝑑𝑒𝑡 ∗ 𝑑𝑎𝑦𝑠

𝑈𝑛𝑎𝑣𝑎𝑖𝑙𝑎𝑏𝑖𝑙𝑖𝑡𝑦 = = = = 1.04%
𝑡ℎ𝑒𝑜𝑟𝑒𝑡𝑖𝑐𝑎𝑙 𝑑𝑒𝑡𝑒𝑐𝑡𝑜𝑟 𝑡𝑖𝑚𝑒 σ𝑛𝑖=1 𝑡𝑖𝑡𝑜𝑡𝑎𝑙 2416782 𝑑𝑒𝑡 ∗ 𝑑𝑎𝑦𝑠

Operational Hours per Detector per Year

91
𝐴𝑣𝑎𝑖𝑙𝑎𝑏𝑖𝑙𝑖𝑡𝑦 = 1 − 𝑈𝑛𝑎𝑣𝑎𝑖𝑙𝑎𝑏𝑖𝑙𝑖𝑡𝑦 = 98.96%

8669

In operational mode Disabled due to IS37

29/06/2020 19
Prototype Calculation Tool

• Python script running in CERNs

SWAN environment using Jupyter
Output Fault Tree:
Notebook
• Failure probability
• Operational data are extracted as a function of time
manually from the Infor database
and has to be insert as a fix value
• Contains the calculation formulas
and logic of the fault tree model
and the availability calculation
• Provides the failure probability as a Model Parameter Input:
function of time and the system • Component type
availability • Component amount
• Mean repair time

Result: Availability

29/06/2020 20
Opportunities and Limitations
Reliability Calculation Framework

Pros Cons
• Full integrated approach (calculation model fits to • Operational data are only as good as it is maintained
the failure tracking structure)
• A random failure occurrence is assumed → data set
• High applicability of the results because of the use of needs to be larger to estimate the actual failure
CERN’s “own” operational data occurrence behavior
• Component failures are relatively easy to track in the • All possible failures must be covered by the problem
Infor database codes
• Quality of the available data will increase in the • Small data set until now due to the short time period
future → result applicability will increase as well since the introduction of the closing codes

29/06/2020 22
Detector Unavailability due to IS37
 Detector (un)availability due to IS37 disabling in 2019

𝑑𝑖𝑠𝑎𝑏𝑙𝑒𝑑 𝑑𝑒𝑡𝑒𝑐𝑡𝑜𝑟 𝑡𝑖𝑚𝑒 σ𝑛𝑖=1 𝑡𝑖𝑆𝑇37 25164 𝑑𝑒𝑡 ∗ 𝑑𝑎𝑦𝑠

𝑈𝑛𝑎𝑣𝑎𝑖𝑙𝑎𝑏𝑖𝑙𝑖𝑡𝑦 = = = = 1.04%
𝑡ℎ𝑒𝑜𝑟𝑒𝑡𝑖𝑐𝑎𝑙 𝑑𝑒𝑡𝑒𝑐𝑡𝑜𝑟 𝑡𝑖𝑚𝑒 σ𝑛𝑖=1 𝑡𝑖𝑡𝑜𝑡𝑎𝑙 2416782 𝑑𝑒𝑡 ∗ 𝑑𝑎𝑦𝑠

𝐴𝑣𝑎𝑖𝑙𝑎𝑏𝑖𝑙𝑖𝑡𝑦 = 1 − 𝑈𝑛𝑎𝑣𝑎𝑖𝑙𝑎𝑏𝑖𝑙𝑖𝑡𝑦 = 98.96%

29/06/2020 25
Considered Reliability Data Sources
Reliability Data – Different Sources

Operational Data Data from Manufacturer Literature Data

• High applicability because they • Reliability data are commonly not • Purchasable datasets
are recorded under actual tracked by the manufacturer
conditions (Securiton) • Environmental conditions are not
well documented or to different
• Database is already available • If data is available, no
(Infor) background information can be • Components are not the same →
provided applicability is low
• Standardized failure tracking
codes are available (EDMS
NO.1706394)

High Data applicability Low

29/06/2020 27
Calculation
Fault Tree Model/ Block Diagram
 Calculation Model – Block Diagram Layout
General failure probability:
𝐹 𝑡 = 1 − 𝑒 −𝜆∙𝑡
Negative logic:
OR ... 𝐹 𝑡 = 1 − ς𝑛
𝑖=1(1 − 𝐹𝑖 (𝑡))

AND ... 𝐹 𝑡 = ς𝑛
𝑖=1 𝐹𝑖 (𝑡)

Input 𝑛𝑇𝑦𝑝𝑒 𝑛𝑇𝑦𝑝𝑒 𝑛𝐶𝑜𝑚𝑝 𝑛𝐶𝑜𝑚𝑝𝑃𝐶

Output
Component_Type1 ∙∙∙ Component_Typen 𝐹𝑆𝑦𝑠𝑡𝑒𝑚 = 1 − ෑ 1 − 𝐹𝐶𝑜𝑚𝑝𝑇𝑦𝑝𝑒 𝑖 = 1 − ෑ 1− ෑ 1− ෑ 1 − 𝐹𝐶𝑜𝑚𝑝𝑃𝐶 𝑖
𝑖=1 𝑖=1 𝑖=1 𝑖=1

Component1
𝑛𝐶𝑜𝑚𝑝
∙∙∙

Failure probability of all components: 𝐹𝐶𝑜𝑚𝑝𝑇𝑦𝑝𝑒 = ෑ 𝐹𝐶𝑜𝑚𝑝 𝑖

𝑖=1
Componentn

𝑛𝐶𝑜𝑚𝑝𝑃𝐶

PC1 ∙∙∙ PCn Failure probability of one component:

𝐹𝐶𝑜𝑚𝑝 = 1 − ෑ (1 − 𝐹𝐶𝑜𝑚𝑝𝑃𝐶 𝑖)
𝑖=1

29/06/2020 29
 Calculation Model – Block Diagram Layout
General failure probability:
Legend:
CP ... Control Panel 𝐹 𝑡 = 1 − 𝑒 −𝜆∙𝑡
OPD ... Optical Point Detector
Negative logic:
ASD ... Aspirating Smoke Detector
VSD ... Venturi Smoke Detector OR ... 𝐹 𝑡 = 1 − ς𝑛
𝑖=1(1 − 𝐹𝑖 (𝑡))
AS ... Alarm Siren
PC ... Problem Code (e.g. SP008) AND ... 𝐹 𝑡 = ς𝑛
𝑖=1 𝐹𝑖 (𝑡)

Input Output
CPType OPDType ASDType VSDType ASType 𝐹(𝑡)𝑆𝑦𝑠𝑡𝑒𝑚 = 1 − 1 − 𝐹𝐶𝑃𝑇𝑦𝑝𝑒 1 − 𝐹𝑂𝑃𝐷𝑇𝑦𝑝𝑒 1 − 𝐹𝐴𝑆𝐷𝑇𝑦𝑝𝑒 1 − 𝐹𝑉𝑆𝐷𝑇𝑦𝑝𝑒 1 − 𝐹𝐴𝑆𝑇𝑦𝑝𝑒

OPD1
𝑛𝑂𝑃𝐷
∙∙∙

Failure probability of all optical point detectors: 𝐹𝑂𝑃𝐷𝑇𝑦𝑝𝑒 = ෑ 𝐹𝑂𝑃𝐷 𝑖

𝑖=1
OPDn

𝑛𝑃𝐶−𝑂𝑃𝐷

PC1 ∙∙∙ PCn Failure probability of one optical point detector:

𝐹𝑂𝑃𝐷 = 1 − ෑ (1 − 𝐹𝑃𝐶 𝑖)
𝑖=1

29/06/2020 30
Calculation
Operartonal Data
Operational Data

Operational time:
# obj_code
• Counting since the introduction of closing codes
• New installed (not renewed) components are not SFDEI-19xxx
SFDEI-19xxx
taken into account SFDEI-19xxx
SFDEI-19xxx
• Downtime due to repair/ renewal is neglected
(because max. 2h per failure)
.
.
.
Failure counting:
• Failures are tracked by work orders with problem Introduction of Today
codes closing codes Time

• Different problem codes are counted independently

• Counting since the introduction of closing codes

29/06/2020 32
Reliability Parameter Estimation

Weibull analysis:
• In reliability engineering commonly used to evaluate
reliability data
• It is flexible to apply and it can cover different failure rates
(bathtub curve)

2 parametric Weibull distribution:

Probability distribution function (pdf) Failure rate
𝑏−1 𝑏
𝑏 𝑡 𝑏−1
𝑏 𝑡 𝑡
− 𝑇
𝑓(𝑡) = ∙ ∙ 𝑒 𝜆(𝑡) = ∙ b <1 b =1 b>1
𝑇 𝑇 𝑇 𝑇
Where: λ(t) = const.
b … slope/ shape parameter
t … failure time
T … characteristic life/ scale parameter

29/06/2020 33
Reliability Parameter Estimation

Assumptions on the data

• Failures occur randomly and time independent → λ(t) = const. → b = 1

Weibull function Exponential function

𝑏 𝑡 𝑏−1 𝑡 𝑏 b=1 1 −1∙𝑡
𝑓(𝑡) = ∙ ∙ 𝑒
− 𝑇 𝑓(𝑡) = ∙ 𝑒 𝑇 = 𝜆 ∙ 𝑒 −𝜆∙𝑡
𝑇 𝑇 𝑇
Expected MTBF
1 σ𝑛𝑖=1 𝑡𝑖
𝐸 𝑋 = = 𝑀𝑇𝐵𝐹 = 𝑛
𝜆 σ𝑖=1 𝑟𝑖
Weibull failure rate Exponential failure rate
𝑏 𝑡 𝑏−1 b=1 1 1
𝜆(𝑡) = ∙ 𝜆(𝑡) = 𝑇 = 𝑀𝑇𝐵𝐹
𝑇 𝑇

29/06/2020 34
Confidence Interval

Quantifying the uncertainty 60%

99%

• Lifetime data are only estimates due to a limited sample size

Point estimation
• The smaller the failure amount, the bigger the uncertainty of the “true”
MTBF
• Quantify the uncertainty by confidence bounds

→ The confidence limits for exponential distributed values are given by 1 10 100 1000
# failures
the 2 (Chi-square) distribution

2∙𝑡 2∙𝑡
, For e.g.:
𝜒 2 𝛼,2𝑟+2 𝜒2 𝛼
1− 2 ,2𝑟 If the acceptable error risk is assumed with α=10%, the two sided boundaries for the
2 confidence interval are 5% and 95%
Where:
t … total unit hours → In 90% of the times the confidence interval covers the “true” value
r … total numbers of failure
2 … Chi-square distribution value
α … Acceptable risk of error

29/06/2020 35