etrology e a s u r e m e n t S c i e n c e a n d T e c h n o l o g y ETH How Loading Errors Arise in Metrology Lesson Karl H. Ruhm Institute of Machine Tools and Manufacturing (IWF), ETH Zurich, Switzerland ruhm@ethz.ch 08. 03. 2010 Version 04; 24.08.2011 www.mmm.ethz.ch/dok01/d0000933.pdf There exists a German version of this document: d0000765 Aim Sense and understand loading structures of mutual relations and their consequences on measurement quality. ETH 2 Administrative Matters ETH 3 • What is Covered? Overview on Different Fundamental Topics Overview on Interrelations of these Topics (Structures) Theory and Practice Examples and Applications Block Diagrams (Signal Effect Diagrams) for Visualisation • What is not Covered? Recipes for Today and for Everybody "Measurement Uncertainty" (→ GUM) "How to Measure …" Instrumentation Design Product Information ETH 4 Content Administrative Matters ETH 1 Problems and Concepts 2 Measurement Setting – Process und Measurement Process 3 Principle "Loading" – Qualitative Examples 4 Principle "Loading" – Quantitative Examples 5 Generalisation – Back-Loading Structures Summary 5 1 Problems and Concepts ETH 6 Statements • Measurement procedures are never ideal: Nonideal Measurement. • However, the concept of Ideal Measurement is immensely convenient to define and to fight measurement errors and uncertainties. • So, nonideal processes are viewed and judged from the perspective of virtual, ideal processes. ETH Supplement → Module "Ideal Measurement Process" Supplement → Module "Nonideal Measurement Process" 7 Statements There are three, and only three, causes of measurement errors: • nonideal transfer response behaviour of the measurement path (transfer errors) • disturbing quantities and nonideal transfer response behaviour of the disturbance path (disturbance errors) • nonideal transfer response behaviour of the loading path and resulting loading quantities (loading errors) Here, we will treat the last topic only: Loading Errors ETH Supplement → Module "Nonideal Measurement Process" 8 Statements ETH • Properties and behaviour of processes of interest and their quantities may be influenced and disturbed by attached measurement processes. • Within the fields of natural and technological sciences the causes of these disturbances are always of pure physical nature. • Withdrawal of information goes with withdrawal of energy; so called "loading effects", "back-loading effects", "burden effects" will occur. • Unwanted "Loading" can be understood as a generalised "impedance mismatch". 9 Statements ETH • Loading (burden) effects provoke deviations of the quantities to be measured. • Deviations of the quantities to be measured are not visible in the immediate measurement results. • Therefore, we detect loading (burden) errors only seldom. • Whether loading (burden) errors are disturbing or acceptable, is judged individually by means of the independent Quality Assurance Process Q. 10 Prerequisites and Methods – Qualitative ETH • The physical relations within the process without instrumentation must be known. • The physical relations within the measurement process must be known. • The physical interrelations between process and measurement process in the closed loop, due to loading, must be known. • Modelling of the processes is done using concepts of Signal- and System Theory. • The definition of the loading (burden) error is given by Measurement Theory. 11 2 Measurement Setting – Process und Measurement Process ETH 12 Methods – Qualitative Starting point: "Ideal Circumstances" • Model of the ideal process P • Model of the ideal measurement process M • Model of the ideal series connection of process and measurement process • No disturbing feedback effects at all quantities actually measured y(t) u(t) ! process P yˆ (t) − y(t) = o B1124 process domain instrumental process domain measurement process M process under measurement PUM ETH y(t) ˆ resulting quantities Supplement → Module "Ideal Measurement Process" 13 The ideal structure of measurement is described by an ideal series connection: only feed-forward acting quantities B1356 u(t) = u 2(t) y 1(t) = y(t) y 2 (t) = u 1(t) 2 1 → no feed-back quantities, no loading quantities ← This concept • is useful as a working hypothesis • is often far from reality and inapplicable • has to be improved ETH Supplement → Module "Group-Notation of Signals" Supplement → Module "Elements of the Signal Effect Diagram" 14 The ideal structure of measurement is described by an ideal series connection: only feed-forward acting quantities Example, approximately ideal: Principle of Optical Distance, Velocity, Vibration Measurement Figure: Polytec ETH → no feed-back quantities, no loading quantities ← 15 The ideal structure of measurement is described by an ideal series connection: only feed-forward acting quantities Example, series connection of process P and measurement process M approximately ideal: Principle of Optical Stress and Strain Measurement Figure: Polytec → no feed-back quantities, no loading quantities ← ETH Supplement → External Document "Strain Measurement" 16 Methods – Qualitative Reality: "Nonideal Circumstances" Example series connection of three "nonideal" processes: Principle of Piezoelectric Sensor, Cable and Charge-to-Voltage Converter ETH Figure: Kistler 17 Methods – Qualitative Reality: "Nonideal Circumstances" • Model of the nonideal process P • Model of the nonideal measurement process M • Model of the series connection with feedback-effects → closed loop connections involved ← ↓ We have to investigate those structures in order to understand the relevant relations! ↓ ETH Supplement → Module "Nonideal Measurement Process" 18 3 Principle "Loading" – Qualitative Examples ETH 19 General Principle "Loading" anywhere in • Natural Sciences and Technology • Business Administration, Economics, Sociology, Psychology • Community not only in Metrology ETH 20 The ideal structure of measurement is described by an ideal series connection: only feed-forward effecting quantities B1356 u(t) = u 2(t) y 1(t) = y(t) y 2 (t) = u 1(t) 2 1 → no feed-back quantities, no loading quantities ← Nondynamic Transfer Response Equation y(t) = g1 g2 u(t) [{y}] with gi [{yi ui−1}] static transfer response values Supplement → Module "Group-Notation of Signals" ETH Supplement → Module "Elements of the Signal Effect Diagram" Supplement → Module "Concept and Term «Transfer» Concerning Systems" 21 nonideal series connection: feed-back effecting quantities • Extended series connection • Series of interlaced loop connections Example 1 Electric current i(t) through a RC-Process B1354 source ETH Δu 1 Δu 2 Δu 3 Δu 4 R1 R2 R3 R4 u0 (t) C1 C2 C3 C4 u(t) sink i(t) 22 nonideal series connection: feed-back effecting quantities • Extended series connection • Series of interlaced loop connections Example 2 * through the layers of a wall Heat current Q(t) R1 ϑ0 (t) ETH R3 R4 * Q(t) B1355 source R2 C1 C2 Δu 1 Δu2 C3 C4 ϑ(t) sink Δu 3 Δu 4 23 Example 3 Extended series connection of multiple sub-processes Process Diagram tank T power electronics E uE(t) i (t) M M = M motor M * T(t) m p (t) A(t) cons gearbox GB ω (t) GB * V(t) m pC(t) compressor C B0630 u(t) ω (t) valve V How does the Signal Effect Diagram look like? ETH 24 Example 3 Extended series connection of multiple sub-processes Process Diagram tank T power electronics E M = M motor M i (t) * V(t) m pC(t) * T(t) m p (t) A(t) cons gearbox GB ω M compressor C (t) GB B0630 uE(t) u(t) ω (t) valve V Signal Effect Diagram iE(t) ML * (t) m T (t) GB ω M(t) u(t) pcons(t) p (t) C * V(t) m B0629 A(t) ETH uE(t) power electronics E iM(t) motor M ω (t) gearbox GB GB compressor C ML (t) C tank T pT (t) valve V * (t) m V 25 Generalisation Extended series connection → Series of interlaced loop connections B0003 or or ETH 26 Extended series connection → Series of interlaced loop connections basic physical structure currents (matter, energy, pulse, momentum) flow through extended series connections of resistors and storages B1358 special basic system structure quadrupole u 1 (t) y1(t) u1(t) y2 (t) or u (t) u 2 (t) 2 y1(t) y2(t) or u(t) y(t) general basic mathematical structure band matrices B1357 0 ETH 0 Supplement → Module "Generalised Kirchhoff-Relations" Supplement → Module "Principle Resistance" 27 Example simple electric quadrupole Voltage Divider Process Diagram Δu R(t) iR(t) R 1 i(t) u1(t) i1(t) R1 Signal Effect Diagrams detail u(t) u1(t) i(t) i1(t) 1 R1 R process iR (t) i(t) ETH Σ – Δu (t) R B0648 u(t) + u1(t) B0668 global iR (t) B0667 u(t) 1 Σ + i1(t) + 28 Generalised "Quadrupole" for example Bidirectional Closed Information Loop in Automatic Control matter matter energy momentum energy momentum B1227 process P actuation process A process domain instrumental process domain measurement process M control process C information information controlled process CP ETH Remember: All lines represent potential multiple signals, modelled as vectors. 29 Importance of extended series connection • • • • • forward and backward acting quantities exchange of matter, energy, pulse, moment etc. energy and power relevant practice oriented demanding concerning modelling source extensive quantities ETH B0005 intensive quantities drain Supplement → Module "Extensive, Intensive Signals" 30 Examples of Power Demand electrical power source thermal power source ∗ P(t) = c m(t) Δϑ(t) [W] P(t) = i(t) Δu(t) [W] Δϑ(t) P(t) Π B1352 B1352 u(t) Π * = P(t) Q(t) * c m(t) i(t) and so on. Note: The quantity "Power" is always composed of several individual quantities. ETH Supplement → Examples "Physical Power" 31 4 Principle Loading – Quantitative Examples ETH 32 Attempt Nonideal and error related items will be shown in red, ideal items will be shown in green. ETH 33 Phenomenon Loading Quantitative Example 1 Loading of a nonideal voltage source ETH 34 Application Sensors as Voltage Sources Always as First Question How would the ideal loaded voltage source look like? Equivalent Circuit Diagram source u0 iL B1091 Q uL resistance uK RL (load) L i Fundamental Relations (non-dynamic transfer response functions) current output voltage i= 1 u0 RL [A] uK = f(u0 ) = 1⋅ u0 [V] No other resistances (impedances)! ETH Supplement → Module "Loading of a Voltage Source" 35 Equivalent Circuit Diagrams ideal loaded voltage source iL B1091 Q u0 uL resistance uK RL (load) L uQ iQ source Q RQ uL u0 iL B1090 source nonideal loaded voltage source i i Fundamental Relations (non-dynamic transfer response functions) current output voltage resistance uK RL (load) L 1 i= u RL 0 i= [A] uK = f(u0 ) = 1⋅ u0 [V] 1 u RL + RQ 0 uK = f(RQ ;RL ;u0 ) = [A] 1 u RQ 0 1+ RL [V] ≠1 ETH Supplement → Module "Loading of a Voltage Source" 36 Consequences Deviation in the output voltage uK by the voltage drop uQ static transfer response characteristic output voltage uK = f(i) uK /V uQ iQ RQ u0 uL iL B1090 Q u0 uK(i) resistance uK RL (load) L B1081 source uQ = u0 u K = eu (i) i 0 0 i [A] Remark A controlled voltage source would enforce a source resistance (impedance) of RQ = 0 Ω ETH Supplement → Module "Loading of a Voltage Source" 37 uQ iQ RQ source uL u0 iL B1090 Q resistance uK RL (load) L i Next Step for a System Analysis According to System Theory we look for the model of the nonideal process given by a • signal effect diagram of the expected closed loop u + B1444 = u0 [V] – Σ gv gr y = uK [V] g • overall transfer response function y = f(u) = g u • overall transfer response value g = ETH gv 1 + gv gr Supplement → Module "Concept and Term «Transfer» Concerning Systems" 38 uQ iQ RQ source iL B1090 Q u0 uL resistance uK RL (load) L i Model of the Nonideal Process signal effect diagram u0 +_ Σ B1082 uQ RQ i Q = iL ETH 1 u = guL ,u0 u0 RQ 0 1+ RL 1 RL uL load L source Q transfer response function uK = f(u0) = g u0 uK = uL = u K = uL u 0 uQ [V] g uL,u0 transfer response value g guk ,u0 = gv 1 ! = = 1 [−] R 1 + gv gr 1+ Q RL Supplement → Module "Concept and Term «Transfer» Concerning Systems" 39 u0 +_ Σ uQ B1082 u K = uL u 0 uQ RQ source Q i Q = iL 1 RL uL load L g uL,u0 Goal No Voltage Drops No Loading No Loading Error Possible, if Source Impedance RQ = 0 or Load Impedance RL = ∞ What is to be done, if we cannot reach the goal? → Correction within the following Reconstruction Process R ← (which we need anyway) ETH Supplement → Module "Loading of a Voltage Source" 40 Phenomenon Loading Quantitative Example 2 Back-loading error at a voltage divider ETH 41 Application All measurement principles with voltage dividers e.g.: position, distance or angle measurement Measurement Principle 100% R1 B0437 u0 R2 x u 0% R +R =R 1 2 Draw-Wire Sensor (Figures: MICRO-EPSILON) ETH Supplement → Interactive Animation "Potentiometer" 42 ideal source u1 R1 x R2 i2 i u0 static transfer response characteristic u u real drain RL 1.0 B0435 B0436 equivalent circuit diagram 0 R=R +R 1 0.8 iL u RL = R supply potentiometer 2 measurement device oo 0.6 2 1 0.5 e NL 0.4 We may call the error loading error eload(t) with regard to its real cause or nonlinearity error eNL(t) with regard to its appearance. 0.2 0 0 0.2 0.4 0.6 0.8 1 x= R2 R ETH Supplement → Interactive Animation "Potentiometer" 43 Phenomenon Loading Quantitative Example 3 Back-loading error in a thermocouple TC caused by an amplifier V ETH 44 Loading as an error process E attached to a sensor process S ϑ 2 Δϑ = ϑ1 ϑ2 [K] bei ϑ1 > ϑ2 RQ u ˆV u ˆ Δϑ RL amplifier V ϑ 1 thermocouple TC RQ uΔϑ B1020 B1019 Δu B uˆΔϑ ûV RL iL g uˆ ,uˆ V Δϑ Again a voltage divider! Figure: Analog Devices What can be done? Can we reduce the systematic error eload(t)? → Yes ← ETH Supplement → Example "Electrical Loading of a Thermocouple" 45 Loading as an error process E attached to a sensor process S Δu B RQ B1020 uΔϑ uˆΔϑ ûV RL iL g uˆ ,uˆ V Δϑ What would be ideal? → The Nominal Sensor Process SN ← Δϑ Δϑ process domain instrumental process domain uV B1059 nominal sensor process SN – Σ + error at the output of the sensor process e uv ˆV u sensor process S ETH Supplement → Module "Ideal Measurement Process" Supplement → Example "Electrical Loading of a Thermocouple" 46 Development of the nonideal sensor process S Δϑ Δϑ [°C] [°C] process domain instrumental process domain u Δϑ g u,Δϑ [V] nominal thermocouple TCN gu uV V ,u [V] nominal amplifier VN Note: this is just the model of the process and not physical reality! nominal sensor process SN _ + g u,Δϑ u Δϑ +_ RQ B1315 ETH [V] + u ˆ Δϑ [V] ΔuB = e uΔϑ (source Q) _ Δϑ Σ Σ thermocouple TC eu iL [V] Σ eu V [V] u ˆV gu V ,u 1 RL amplifier V (load L) We identify the error process E, positioned in series between the thermocouple TC and the amplifier V (red loop). [V] Cause of loading errors: Voltage divider with poorly matched parameters (parameter mismatch). Ideal would be: RQ = 0 and / or RL = ∞, then the loading path would be tight (closed). [A] nonideal sensor process S Supplement → Module "Nonideal Measurement Process" Supplement → Example "Electrical Loading of a Thermocouple" 47 Details concerning the Error Process E uΔϑ + Σ _ Δu L = e uΔϑ [V] [V] [V] RQ iL [A] 1 RL eu ↓ B1447 Δϑ [V] error process E The transfer response function of the error process E is 1 ûΔϑ = gE uΔϑ = u RQ Δϑ 1+ RL ETH numerical example RQ = 1.506 Ω RL = 1046 Ω u ˆ Δϑ [V] gE = 0.9985 [V / V] The error concerning the output voltage of the thermocouple TC is euΔQ RQ uΔϑ = RQ + RL Supplement → Module "Nonideal Measurement Process" Supplement → Example "Electrical Loading of a Thermocouple" [V] 48 Δϑ Δϑ [°C] [°C] process domain instrumental process domain u Δϑ g u,Δϑ gu [V] nominal thermocouple TCN We know already, ideal would be: uV V ,u [V] • RQ = 0 and / or • RL = ∞ nominal amplifier VN nominal sensor process SN _ + g u,Δϑ u Δϑ +_ RQ B1315 [V] + gu [V] iL V ,u 1 RL [V] amplifier V [A] nonideal sensor process S ETH Σ eu then the loading path would be closed. V [V] But, that's impossible: u ˆV u ˆ Δϑ ΔuB = e uΔϑ (source Q) _ Δϑ Σ Σ thermocouple TC eu (load L) [V] We have to compensate error sources of the sensor process S within the following reconstruction process R. Supplement → Module "Nonideal Measurement Process" Supplement → Example "Electrical Loading of a Thermocouple" 49 Any Sensor Process S needs a Reconstruction Process R (Fundamental Axiom of Metrology). B0398 ideal • Sensor Process S (mapping) pressure sensor • Reconstruction Process R (inversion) 7.52 bar ˆ p/bar 2.71Volt u p /V Example p/bar actually measured quantity y(t) u(t) process P Σ B1406 process domain instrumental process domain + ! yˆ (t) − y(t) = o sensor quantity y(t) = u S(t) OpS {...} yS(t) = uR(t) 1 OpS {...} yR (t) sensor reconstruction process S process R measurement process M ETH error quantity e y(t) = y(t) ˆ erroneous resulting quantity Supplement → Module "Sensor Process" Supplement → Module "Reconstruction Process – A Survey" 50 Any Sensor Process S needs a Reconstruction Process R. nonideal • Sensor Process S (mapping) • Error Process E (inference) within the sensor process S in a series connection • Reconstruction Process R (inversion of mapping and inference) y(t) B1454 OpS {...} ETH y S (t) sensor process S OpE {...} yˆ S (t) 1 OpE {...} yScorr(t) 1 OpS {...} ˆ y(t) reconstruction process R measurement process M Supplement → Module "Sensor Process" Supplement → Module "Reconstruction Process – A Survey" Supplement → Module "Inversion of Interconnected Systems" 51 The usual incomplete reconstruction process R at a loaded sensor process S Δϑ Δϑ [°C] process domain instrumentation domain [°C] u Δϑ g u,Δϑ [V] nominal thermocouple TCN gu uV V ,u [V] nominal amplifier VN nominal sensor process SN _ + g u,Δϑ u Δϑ RQ +_ (source Q) _ [V] + Δϑ Σ u ˆ Δϑ Σ [V] ΔuB = e uΔϑ thermocouple TC eu [V] Σ eu _ V [V] u ˆV gu V ,u [V] (load L) [A] B1018 1 gu ,u V [°C] u ˆ Δϑ 1 ˆ Δϑ [V] g u,Δϑ [°C] 1 RL amplifier V iL + Σ eΔϑ nonideal sensor process S incomplete invers model of amplifier V incomplete invers model of thermocouple TC incomplete reconstruction process R nonideal measurement process M ETH Supplement → Module "Reconstruction Process – A Survey" Supplement → Example "Electrical Loading of a Thermocouple" 52 Possibilities shown by the Transfer Response Function Ideal Measurement Process Δϑˆ = 1 1 gu,Δϑ guV ,u guV ,u gu,Δϑ = 1⋅ Δϑ [°C] Incomplete Reconstruction Process R Δϑˆ = 1 1 gu,Δϑ guV ,u guV ,ugEgu,ΔϑΔϑ = gEΔϑ [°C] = Complete Reconstruction Process P Δϑˆ = 1 gu,Δϑ 1 1 guV ,ugEgu,ΔϑΔϑ = 1⋅ Δϑ [°C] gE guV ,u The usual incomplete reconstruction process R at a loaded sensor process S leads to loading errors eDJ. ETH Supplement → Module "Reconstruction Process – A Survey" Supplement → Example "Electrical Loading of a Thermocouple" 53 The complete reconstruction process R at a loaded sensor process S (Assumption: Ideal Reconstruction Process R; no numerical errors) Δϑ Δϑ [°C] [°C] process domain instrumental process domain u Δϑ g u,Δϑ gu [V] nominal thermocouple TCN uV V ,u [V] nominal amplifier VN nominal sensor process SN _ + g u Δϑ u,Δϑ RQ +_ B1017 _ [V] + Δϑ Σ u ˆ Δϑ Σ gu [V] Δu L = e uΔϑ thermocouple TC (source Q) eu iL V ,u amplifier V (load L) _ V [V] u ˆV 1 RL [V] Σ eu [V] + u ˆ Δϑ 1 g u V ,u [V] 1 RL + ûΔϑcor 1 [V] gu,Δϑ Σ + Δu L = e uΔϑ invers model of amplifier V [A] ˆi L [V] Σ eΔϑ= 0 [°C] Δϑ [°C] RQ invers model of thermocouple TC [A] nonideal sensor process S reconstruction process R measurement process M ETH Supplement → Module "Inversion of Interconnected Systems" Supplement → Module "Reconstruction Process – A Survey" 54 The complete reconstruction process R at a loaded sensor process S Example Parameter from a Calibration Process CP transfer response value thermocouple TC: gu,Δϑ = 52.8 μV °C−1 transfer response value amplifier V: gu ,u = 0.992mV μV −1 V RQ = 1.506 Ω source resistance thermocouple TC: RL = 1004 Ω load resistance of amplifier V: Transfer Response Value gS of the Sensor Process S: gS = guV ,u 1 1 + RQ 1 RL gu,Δϑ = 52.299 [mV °C−1] Transfer Response Value gR of the Reconstruction Process R: gR = ETH 1 gu,Δϑ (1 + RQ 1 1 = 0.01912 [°C mV −1] ) RL guV ,u Supplement → Module "Inversion of Interconnected Systems" Supplement → Module "Reconstruction Process – Survey" 55 Intermediate Results • Unmatched parameters within measurement processes M lead to systematic loading errors eload(t). • Loading effects within a measurement process can be identified by a calibration procedure of the measurement process (modelling). • Loading effects within a measurement process M can be systematically compensated by a complete reconstruction R process. • For the example at hand the back-loading error eload(t) is available on demand, since all information is known. Unfortunately, this is not always the case. ETH 56 Prerequisites for these results: knowledge (model) about process P and sensor process S: → Model Based Measurement ← Assumptions and Constraints for these results: • • • • linear behaviour of all elements and signal interactions nondynamic behaviour of all elements no deterministic or random disturbance signals ideal behaviour of the reconstruction process R (no numerical errors) → More or Less Unrealisable in Practice ← But doing nothing is worse! ETH 57 Further Things to Do • Dynamic Sensor Process S • Dynamic Back-Loading • Dynamic Reconstruction • Impedance Matching • Dynamic Process P • Extended Reconstruction Process Some Examples and Results ETH 58 Phenomenon Loading Quantitative Example 4 Dynamic thermal loading by a temperature sensor ETH 59 Again, loading acts as a disturbance quantity at the process approximate signal effect diagram process diagram R(t) [Ω] * Q23(t) 3 ϑ (t) 1 3 ϑ 1 (t) environment 3 B0747 ϑ1(t) 3 ϑ 2 (t) m 3 = oo ; ϑ3 ; QSp * (t) Q 13 S ϑ2(t) IC u(t) 2 Pel(t) * (t) Q 21 2 m 1 ; ϑ 1 ; Q Sp 1 R(t) 1 sensor process process B0745 P(t) [W] m2 ; ϑ2 ; QSp ϑ 2 (t) 2 Fundamental Relations storage equations QSp (t) = mc Δϑ(t) ETH current equations [J] Q(t) = α A Δϑ(t) [Js −1] * Supplement → Exercise "Thermal Back-Loading" 60 Again, loading acts as a disturbance quantity at the process detailed signal effect diagram α1A13 – Σ * (t) Q 13 Σ + αA 1 13 d Q (t) dt Sp1 – + QSp (t) 1 sensor process S 1 1 m 1 c1 ϑ 1(t) QSp (0) 1 αA – * (t) Q 21 2 21 Σ + α2A21 ϑ3 environment ETH + αA 3 23 * (t) Q 23 – Σ Σ – 2 QSp (t) 2 2 – + 2 αA 1 m 2 c2 ϑ 2 (t) B0748 Pel (t) QSp (0) process P d dt QSp (t) 3 23 Supplement → Exercise "Thermal Back-Loading" 61 Phenomenon Loading Quantitative Example 5 Dynamic mechanical loading by an oscillatory structure ETH 62 Loading acts as a disturbing quantity at the process hs (t) . h s(t) .. h s(t) h(t) . h(t) .. h(t) m κ c cs F(t) ms B0911 Fs (t) schwingende Struktur κs Sensor single differential equations process P : ma(t) + c(v(t) − v S (t)) + κ(h(t) − hS (t)) = f(t) [N] sensor process S : mS aS (t) + c S (v S (t) − v(t)) + κS (hS (t) − h( t)) = 0 ETH [N] Supplement → Example "Mechanical loading by an oscillatory structure" 63 Loading acts as a disturbing quantity at process P κ m c m v(0) h(0) – – f(t) 1 m 1 m + Σ a(t) v(t) h(t) h(t) v(t) + a(t) process domain instrumentation domain process P κS mS cS + Σ mS – vS (0) + + Σ + Σ – hS(0) a S(t) vS(t) fa (t) S g u ,f u f (t) a f a B1122 mS h S(t) ETH back-load path sensor process S Supplement → Example "Mechanical loading by an oscillatory structure" 64 5 Generalisation – Back-Loading Structures ETH 65 Three Causes of Nonideal Measurement 1. 2. 3. nonideal transfer response (structure, parameter) disturbing quantities and disturbance paths loading quantities and load paths u M(t) B1360 = disturbed measurement quantities ydist(t) vM(t) measurement path Σ measurement / load path yM(t) estimating quantities y(t) ˆ disturbance path disturbance / load path Σ zM(t) loading quantities disturbing quantities measurement process M ETH Supplement → Module "Nonideal Measurement Process" 66 Generalised Fundamental Axiom of Measurement: The transfer response function matrix of the measurement process M is ideal, if firstly, the transfer response function matrix of the measurement path equals the unit matrix I: yˆ (t) = yM (t) = I uM (t) = I y(t) u M(t) = disturbed measurement quantities ydist(t) ! =I ! =0 B1164 ! =0 vM(t) ! =0 measurement path measurement / load path estimating quantities y(t) ˆ disturbance path disturbance / load path disturbing quantities ETH Σ yM(t) zM(t) Σ loading quantities measurement process M Supplement → Module "Nonideal Measurement Process" 67 Generalised Fundamental Axiom of Measurement: The transfer response function matrix of the measurement process M is ideal, if secondly, the transfer response function matrix of the disturbance path equals the zero matrix 0: yˆ (t) = yM (t) = 0 vM (t) u M(t) = disturbed measurement quantities ydist(t) ! =I ! =0 B1164 ! =0 vM(t) ! =0 measurement path Σ measurement / load path yM(t) estimating quantities y(t) ˆ disturbance path disturbance / load path disturbing quantities zM(t) Σ loading quantities measurement process M ETH Supplement → Module "Nonideal Measurement Process" 68 Generalised Fundamental Axiom of Measurement: The transfer response function matrix of the measurement process M is ideal, if thirdly, the transfer response function matrixes of the loading paths equal each the zero matrix 0: zM (t) = 0 uM (t) und u M(t) = disturbed measurement quantities ydist(t) ! =I ! =0 B1164 ! =0 vM(t) ! =0 zM (t) = 0 vM (t) measurement path Σ measurement / load path yM(t) estimating quantities y(t) ˆ disturbance path disturbance / load path disturbing quantities zM(t) Σ loading quantities measurement process M ETH Supplement → Module "Nonideal Measurement Process" 69 y nom(t) not measurable nominal process PN (open loop observer, OLO) (unequipped process) measurement quantities of interest – + Σ ey (t) nom error quantities due to nonideal interrelations nominal domain process domain Nonideal Measurement Process M • concerning nonideal (dynamic) transfer response behaviour • concerning disturbance quantities • concerning loading quantities u(t) y dist(t) v(t) unknown w(t) measurement quantities, disturbed process P process domain instrumental process domain – B1128 + uM(t) yM(t) v M(t) zM(t) Σ nonideal measurement process M ETH e y (t) error quantities due to nonideal measurement = ŷ(t) Two types of measurement errors: 1. error quantities due to nonideal measurement 2. error quantities due to back-loading resulting quantities, erroneous, estimating Supplement → Module "Nonideal Measurement Process" 70 Generalised Model described by a set of equations of the classical System Theory (State Space Description) non-dynamic subsystem D x(0) dynamic subsystem u(t) (M) B Σ . x(t) x(t) (N) (N) C y(t) Σ (P) x (t) = A x(t) + Bu(t) x(0) = x 0 y(t) = C x(t) + Du(t) B0080 A dynamic system u(t) Reduction: (M) A B C D u(t) (M) ETH y(t) (P) y(t) P (P) ⎡ x (t)⎤ ⎡ A B ⎤ ⎡ x(t)⎤ ⎢ y(t) ⎥ = ⎢ C D ⎥ ⎢ u(t) ⎥ ⎥⎦ ⎣ ⎣ ⎦ ⎢⎣ ⎦ ⎡ x (t)⎤ ⎡ x(t)⎤ ⎢ y(t) ⎥ = P ⎢ u(t) ⎥ ⎣ ⎦ ⎣ ⎦ Supplement → Module "Basic Structures of State Space Representation" 71 Detailed model in state space representation (time domain) with disturbances, back-loading and nonideal transfer response D u(t) B input quantities C x(0) x(t) . B1230 x(t) y(t) Σ output quantities F ⎡ x (t)⎤ ⎡ A B E⎤ ⎡ x(t)⎤ ⎡ x(t)⎤ ⎢ y(t) ⎥ = ⎢ C D F ⎥ ⎢ u(t) ⎥ = P ⎢ u(t) ⎥ ⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥ ⎢⎣ z(t) ⎥⎦ ⎢⎣ G H J ⎥⎦ ⎢⎣ v(t) ⎥⎦ ⎢⎣ v(t) ⎥⎦ A H disturbing quantities ETH E G J y(t) = C x(t) + Du(t) + F v(t) z(t) = G x(t) + Hu(t) + Jv(t) Σ v(t) x (t) = A x(t) + Bu(t) + E v(t) Σ z(t) back-loading quantities non-ideal, dynamic, disturbed and back-loading system Supplement → Module "Basic Structures of State Space Representation" 72 Generalisation 1. u P (t) y P(t) B1161 vP(t) PP 2. u P (t) z P(t) vP(t) PP y P(t) process P process P process domain instrumentation domain process domain instrumentation domain uM(t) vM(t) zM(t) yM(t) PM zM(t) u1(t) y1(t) PP uM(t) PM vM(t) nonideal measurement process M 3. z P(t) yM(t) nonideal measurement process M 4. u1(t) y1(t) u (t) S {PP , PM } y (t) 2 2 u2(t) ETH PM y2(t) "P" = parameter matrix observed process OP described by Redheffer-Star-Product Matrix S 73 Special Series Connection in State Space Description: u1(t) Multiplication of two parameter matrices PM und PP in the so-called y1(t) Redheffer-Star-Product Matrix S B1162 u2(t) S {PP , PM } y2(t) observed process OP described by Redheffer-Star-Product-Matrix S Reconstruction of Process und Measurement Process: Inversion of the Redheffer-Star-Product Matrix S Matlab-Function "starp" ETH 74 Summary ETH 75 ETH • Measurement processes load processes – loading errors arise • Loading errors depend on properties of the measurement process as well as on properties of the process – impedance matching task • Loading errors differ structurally from all other measurement errors – loading errors appear in the surroundings of the process • Loading errors are only seldom recognised – loading errors are not discernable by conventional calibration processes • Loading errors have to be considered and errors have to be analysed and discussed concerning specific demands – task for quality assurance • Loading errors can be avoided by load-free series connections – demanding task for theory and practice 76