Structural Health Monitoring: stato dell arte e sviluppi futuri

Transcript

1 Structural Health Monitoring: stato dell arte e sviluppi futuri ALESSANDRO DE STEFANO, Politecnico di Torino ANTONINO QUATTRONE, Politecnico di Torino Gli autori ringraziano per la fruttuosa collaborazione Emiliano Matta, libero professionista, e Gianluca Ruocci, emigrato all Ecole de ponts et chaussées di Parigi, brillanti ricercatori che l Università italiana non sa trattenere.

2 Structural Health Monitoring: What is? What to look for? How? What is?: observation and measurement programme making sense only if strictly related to ordinary maintenance, residual life and safety assessment, decision making support about critical maintenance actions PAY ATTENTION! Monitoring is often associated to damage detection. In fact not every damage is relevant to residual life, not every damage can be easily detected by monitoring actions! NEED OF RISK ANALYSIS!

3 What to look for and How? WG 1, ISHMII-CSHM 1, Waikiki, Honolulu, 2004 EVENTS Strain Deformation Acceleration Temperature MEASURABLE ENTITIES Geo- Ima-ge metry Electric potenti al Acoustic emission or attenuation Chemicals, including moisture Magnetic properties Research needed Fire FAIR FAIR POOR GOOD 1 2 FAIR Y Explosion GOOD FAIR 1 High priority Collision to Girders and columns FAIR POOR 1 POOR 2 Y Earthquake FAIR 2 POOR 1 GOOD FAIR Y Scour 3 4 High priority Traffic loads 6, 5 5 GOOD GOOD 7 Especially WIM Wind GOOD FAIR GOOD GOOD Y Corrosion FAIR FAIR POOR Structural fatigue FAIR High priority for corrosion of prestressing tendons, current methods make indirect measurements only Y for fatigue of bridge deck slabs Dead Load GOOD GOOD GOOD Notes: 1. Important, but difficult to measure 6. Weighing-in-motion (WIM) 2. Current method too tedious 7. Laser scanner can be used to detect change in geometry due to traffic 3. Change in column strains can detect effects of scour 8. Forces in cables of cable stayed bridges have been obtained from 4. Ultrasonic Structural imaging Health has Monitoring: been used stato to map dell arte erosion e due sviluppi to scour futuri A. DE STEFANO their vibration & A. QUATTRONE, characteristics Politecnico di Torino 5. Laser vibrometer and tiltmeters can be used to monitor traffic

4 RISK ANALYSIS Structural assessment how reliable the existing structure is to carry current and future loads and to fulfill its task for a given time period? UNCERTAINTIES (structural, epistemic, social behavior related) Structural models Deterioration mechanisms Material resistances Geometries Measurements error Loads VARING IN TIME Stochastic approaches

5 RISK ANALYSIS (linerized simplified model) Risk= Prob[p S <p R ] Probability function of loading model Probability function of resistance model FAILURE Risk can be seen as the convolution between Hazard and Vulnerability Hazard: probability that in a time t an external (like earthquakes, floods..) or internal (material degradation, fatigue..) event capable of causing damage occurs Vulnerability : Conditional probability that, when an event occurred, the whole structure or a part of it suffers a predefined damage

6 Approach to safety: from variable time to symptom Reliability based on time: R( t) P( t t ) 1 b RISK Probability that the time it takes a system to reach a damage limit state associated to a damage admissible level, t b, is greater than a generic time t Symptoms can be regarded as evolutionary and sudden changes in observable qualitative properties and/or measurable responses. Correlating symptoms to damage can require a knowledge based direct search or direct incomplete knowledge supported by a model based predictive assessment. Reliability based on symptom: R( S) P( S S b S S l ) fs ds Probability that a system, which is still able to meet the requirements S for which it has been designed (S<S l ), is active and displays a value of the S smaller than S b

7 EFFECTS OF MAINTENANCE ON STRUCTURAL PERFORMANCE Lifetime structural performance without maintenance and with maintenance (1) Preventive maintenance Increase in performance Decrease in the rate of deterioration 1) from Thomas B. Messervey, Integration of Structural Health Monitoring into the Design, Assessment, and Management of Civil Infrastructure, Ph.D. Thesis

8 CONDITION AND RELIABILITY DETERIORATION The condition deterioration profile refers to deterioration in VISUAL terms of singular components of the structure. deterioration occurs at discrete intervals using a stochastic process based on historical records The reliability deterioration profile refers to deterioration of a measure of structural performance, defined by b Combined effects of different failure modes can be capture by reliability deterioration profiles Reliability Index (t=0 Overdesign) b R 2 R S 2 S μ R, μ S :mean values resistance and load effect σ R,σ S :SD of resistance and load effect

9 In general, the capacity (resistance) of a structure decreases over time as the structure deteriorates and the load demand increases. Reaching unacceptable performance (or collapse) during the operational lifetime Time of Failure GOAL: Prediction of a realistic life cycle performance of the structure and his singular components (deterioration models ) Define effective maintenance and risk mitigation programs

10 Reliability of the monitored system R 0 (S) primary reliability that applies to a given type of systems; R(S,L) reliability characterised for the particular system by the introduction of a logistic vector L i ; L i denotes the individual element of the sample, it may contain a series of specific parameters depending on which aspect of the system we want to monitor. It is defined: and putting S R( S, L) exp h( x, L) dx 0 h(s, L) = h 0 (S) g(l), (, ) (, ) 1 T g R S L R ln( ( )) 0 S L 0 L R 0 S L 0 L L where g(l) is an unknown function to be defined, in the assumption of small changes of L :

11 Symptom Observation Matrix (SOM) Column 3: set of observations of the symptom 3 Row i: set of symptoms at observation i S(i,1) S(i,2) S(i,3) S(i,N) LOGISTIC VECTOR L i

12 The SVD is an exact decomposition and it leads to optimized orthogonal components; SOM[p,r]=U[p,r]*SV[diag r,r]*v T [r,r] rr pxr rxr diag rxr U p SOM = Unitary matrix (UU H =U H U= I): x SV Scaling factors x V T Unitary matrix (VV H =V H V= I):

13 Important property of SVD Given an integer number s <r the sum s u k xsv k xv T k k=1 supplies the optimal approximation of SOM given the reduced rank s Suppose that s=1 In such case SOM is approximated by a one-dimension vector product: Approx 1 (SOM)[pxr] u 1 [px1]xsv 1 xv 1T [1xr]

14 r rxr diag rxr u 1 (px1) SV 1 v T 1(1xr) p x SV x Approx 1 (SOM) u 1 (j) is not equal but not far the average of the j th row of SOM and can be considered as a a set of values representative of the state of the structure at each j th observation. A set of damage states shall be associated to each u 1 (j) value through a multi-model based exploration. V T 1(i) is not equal but not far from the average of the i th column of SOM and can be considered as a set of representative values of the significance of each i th symptom

15 Given the previously stated optimal approximation property, the difference: SOM Approx 1 (SOM) is a matrix of residuals (or errors) with the minimum possible Frobenius norm (i.e. the minimum sqare error), compatible with any one-dimensional approximation of SOM. In Approx 1 (SOM) the vector v 1, given that the SOM columns are effectively centered and normalized, can be interpreted as average, invariant significance of the symptoms along the observation process, whilst the vector u 1 contains a set of numbers that can be correlated with the damage state of the structure, influenced but not strictly governed by the scattering and evolution of symptoms. Approx 1 (SOM) leads to the best first order linearized assessment of the damage evolution and symptom significance.

16 Once removed the shadow of the first dominating components, the residual matrix can bear indirect information on defects or damages. The u 2 and v 2 vectors of the SOM are also the first order orthonormal expansions of the residual matrix of the first order approximation The SV j s associate to each corresponding couple of singular vectors a weight (scaling factor); It is reasonable to expect that as much their values are lesser than SV 1 so much better and more robust is the first order assessment and less relevant the contribution of higher order components.

17 Life-Cycle Assessment Civil structures have long service live. During this period are subjected to: DISCRETE TIME Abnormal loadings: earthquakes floods very strong winds fire vehicle impact CONTINUOS TIME Exposure to aggressive environmental stressors: heating/cooling cycles loading cycles increasing of loads chloride attack Lifetime REDUCTION OF THE CAPACITY OF SAFETY CARRY LOADS

18 STEPS TO THE EVALUATION OF STRUCTURAL PERFORMANCE Estimate the range of lifetime Evaluate the deterioration model in service lifetime Sources of impact affecting the structural performance: loads effects impact of increasing loads level (i.e. traffic loads on bridges) environmental influences (temperature, radiation, frost action) degradation due to chemical exposure Inclusion of discrete time events (expectation model and real observation) Assessment criteria of real degradation progress Structural Health Monitoring (SHM)

19 MONITORING STRATEGIES PERIODIC VISUAL INSPECTION WITH DESCRIPTION OF CONDITION PERIODIC VISUAL INSPECTION WITH CLASSIFICATION OF CONDITION IN TERMS OF DEGRADE AND VULNERABILITY PERIODIC VISUAL INSPECTION WITH AMBIENTAL DYNAMIC TESTS PERMANENT MONITORING AND PERIODIC VISUAL INSPECTION ON-LINE MONITORING WITH INTEGRATED RISK ANALYSIS AND DIRECT INSPECTION IN WARNING CASES Actually, most of existing bridge management programs (BRIDGIT, PONTIS in USA), are almost exclusively based on visual inspections (second approach).

20 STRUCTURAL HEALTH MONITORING (SHM) PERMANENT MONITORING Static monitoring displacements cracks opening chemical exposure Pressure Dynamic monitoring Accelerations (modal parameter) Strains Absolute position (GPS, Radar scanner) Temperature, humidity, wind Weigh in motion LOCAL RESPONSE GLOBAL RESPONSE Global dynamic monitoring systems provide useful information to understand the structural behaviour and to detect the damage symptoms

21 Vibration-based SHM approach Aimed at the characterisation of the structural health state via a non-destructive assessment Share the same goals of visual inspections overcoming limitations in a more automatic way Detection of damage symptoms among the features extracted from vibration signatures The damage assessment process can be subdivided into 4 steps: OPERATIONAL EVALUATION Definition of likely damage affecting the structure Definition of the operational condition of the monitoring system and data acquisition limitations Farrar, C. R., Doebling, S. W., Nix, D. A., (2001) Vibration-based structural damage identification, Phil. Trans. R. Soc. A., 359, pp

22 Vibration-based SHM approach Aimed at the characterisation of the structural health state via a non-destructive assessment Share the same goals of visual inspections overcoming limitations in a more automatic way Detection of damage symptoms among the features extracted from vibration signatures The damage assessment process can be subdivided into 4 steps: OPERATIONAL EVALUATION DATA ACQUISITION AND CLEANSING Design of the sensing system (type, number, location of sensors) Definition of the acquisition and sampling frequency, data normalization and noise reduction Farrar, C. R., Doebling, S. W., Nix, D. A., (2001) Vibration-based structural damage identification, Phil. Trans. R. Soc. A., 359, pp

23 Vibration-based SHM approach Aimed at the characterisation of the structural health state via a non-destructive assessment Share the same goals of visual inspections overcoming limitations in a more automatic way Detection of damage symptoms among the features extracted from vibration signatures The damage assessment process can be subdivided into 4 steps: OPERATIONAL EVALUATION DATA ACQUISITION AND CLEANSING FEATURE SELECTION Model/Non-model based sensitivity analysis to identify the best features Condensation of significant information in reduced-size features vectors Farrar, C. R., Doebling, S. W., Nix, D. A., (2001) Vibration-based structural damage identification, Phil. Trans. R. Soc. A., 359, pp

24 Vibration-based SHM approach Aimed at the characterisation of the structural health state via a non-destructive assessment Share the same goals of visual inspections overcoming limitations in a more automatic way Detection of damage symptoms among the features extracted from vibration signatures The damage assessment process can be subdivided into 4 steps: OPERATIONAL EVALUATION DATA ACQUISITION AND CLEANSING FEATURE SELECTION STATISTICAL MODEL DEVELOPMENT Implementation of damage assessment algorithms to detect, localize, classify and quantify damage Testing of the reliability of the developed model in terms of features sensitivity and false indications Farrar, C. R., Doebling, S. W., Nix, D. A., (2001) Vibration-based structural damage identification, Phil. Trans. R. Soc. A., 359, pp

25 STRUCTURAL HEALTH MONITORING (SHM) SHM techniques use the data of monitoring system applying damage detection techniques to track the healthy state of the structure Damage detection levels Data-driven damage detection Detection: Is damage present? Localization: Where is the damage located? Diagnosis: How severe is the damage? Prognosis: What is the remaining safe lifetime? Model-based damage detection

26 Experimental background: accuracy of dynamic identification techniques (OMA) FREQUENCY TIME TIME-FREQUENCY Welch PSD Ewins-Gleeson Dobson Kennedy-Pancu Spectral Multimatrix ARMAV PRTD ERA SSI Wavelets, packet wavelets Time-Frequency Istantaneous Estimator (TFIE) Non stationary input Stability of phase: good for higher damping FDD

27 Data-driven damage assessment Data-driven techniques can be utilized to avoid direct dependence on analytical models. Novelty/outlier analysis Statistical methods Direct interpretation of sympthoms

28 The masonry bridge experimental model Case study of a national research project concerning the surveillance and maintenance of historical structures and infrastructures 1.60m 1.75m 5.90m Global monitoring: vibration tests Application of pier settlement Damage states of increasing extent Model-based damage assessment

29 Politecnico di Torino Dept. of Structural Engineering IL MONITORAGGIO DELLE STRUTTURE NELLA PREVENZIONE DEL RISCHIO SISMICO, ROMA, The damage scenario: scour simulation Settlements application system Hydraulic flume tests Scour profile image monitoring Screws and bearings to introduce differential settlements Numerical simulation and settlements calculation

30 Politecnico di Torino Dept. of Structural Engineering IL MONITORAGGIO DELLE STRUTTURE NELLA PREVENZIONE DEL RISCHIO SISMICO, ROMA, The damage scenario: differential settlements Refer ence DS 0 DS 1 DS 2 DS 3 30cm polystyrene removed 40cm polystyrene removed 60cm polystyrene removed 75cm polystyrene removed 0.5mm settlement applied 1.5mm settlement applied 2.5mm settlement applied Free and forced (hammer impacts) vibration measurements after each damage state

31 Politecnico di Torino Dept. of Structural Engineering IL MONITORAGGIO DELLE STRUTTURE NELLA PREVENZIONE DEL RISCHIO SISMICO, ROMA, The dynamic response tests Sensors locations: vertical on spandrel walls transversal to spandrel walls Experimental setups: longitudinal to the pier transversal to the pier orthogonal to arch barrels 18 monoaxial accelerometers several sensors configurations investigated signals acquired with 400Hz sample frequency

32 Transmissibility Magnitude IL MONITORAGGIO DELLE STRUTTURE NELLA PREVENZIONE DEL RISCHIO SISMICO, ROMA, Outlier Analysis Damage detection algorithm OPERATIONAL EVALUATION DATA ACQUISITION AND CLEANSING FEATURE SELECTION STATISTICAL MODEL DEVELOPMENT PATTERN RECOGNITION: Outlier Analysis D Statistical method which detects novelties as deviations from normal condition The first session data set assumed as reference condition T 1 x x S x x D x novelty detector scalar value (Mahalanobis squared distance MSD) single observation vector TF(ω i ) TF(ω k ) x S reference sample mean vector reference sample covariance matrix TF(ω N ) Spectral lines

33 Mahalanobis Square Distance Mahalanobis Square Distance IL MONITORAGGIO DELLE STRUTTURE NELLA PREVENZIONE DEL RISCHIO SISMICO, ROMA, Outlier Analysis Damage detection algorithm OPERATIONAL EVALUATION DATA ACQUISITION AND CLEANSING FEATURE SELECTION STATISTICAL MODEL DEVELOPMENT PATTERN RECOGNITION: Outlier Analysis TF(ω i ) Outlier Analysis: Natural Frequencies for best NF TF(ω k ) Outlier Analysis Fitness TF(ω N ) samples Outlier Analysis: Damping Ratios for best NF The discordancy value is compared with a statistically computed threshold If the value is greater than the threshold 80 the novelty is detected and damage can be inferred 60 The fitness of each solution is expressed as the area obtained subtracting the threshold value 40 from the series of the Outlier Analysis results and maximised by the genetic optimisation samples

34 Model-based damage assessment Model-based damage assessment methods compare the measured structural response with a numerical simulation generally provided by a FE model The model accuracy is essential to supply a reliable image of the structural health

35 Mismatch between the experimental and the numerical modal parameters The Model Updating Unreliable definition of the reference healthy state of the model Unfeasibility to apply a model-based approach to damage assessment Model updating techniques try to solve the problem but generally deterministic approaches fail because the unique optimal solution they pretend to find is prevented by the inverse nature of the problem. The final result is biased by several errors and uncertainties sources referred to: the experimental measurements the modal identification results the simplified modelling assumptions the construction complexity A single optimal solution is an hard task to accomplish! At list a regularization (e.g.: Tikhonov) is required to reduce the uncertainties, but the regularization is not able to resolve the ambiguities

36 The stochastic Model Updating probabilistic representation of the structural updating parameters The stochastic model updating methods can deal with uncertainties and problem complexity in a robust way multiple model generation driven by the parameters probability the updating output is a class of reliable models selected among all the generated solutions

37 Method outline Sensitivity analysis Parameters ranges estimation Models generation Preliminary models selection definition of the most sensitive parameters Solutions analysis and clustering reduction of the output research space creation of a large models population by an optimisation algorithm exclusion of the less reliable solutions application of data mining techniques to group the selected models

38 Experimental modal analysis + FE model calibration: AN INVERSE PROBLEM Data Model parameters In Civil Engineering, caution is needed even for simple systems to avoid ill-conditioning, especially when ambient vibration is used, and both stiffness and mass parameters are unknown IN THIS WORK results from the JETPACS case study are presented to highlight some crucial robustness issues in vibration-based modelupdating and suggest possible criteria to improve reliability

39 Introduction DPC-ReLUIS Project JETPACS (Joint Experimental Testing on Passive and semi-active Control Systems) 8 Universities involved in the assessment of energy dissipation devices for seismic protection. Control devices A representative FE model is desired, to be shared by all participants for test calibration and interpretation. A preliminary campaign of dynamic tests is conducted, and subsequent attempts, by various partner Research Units, to a parametric identification. Structural Engineering Lab at the University of Basilicata

40 FE model-updating (2/3) In iterative approaches, model-updating = optimization problem, where discrepancies between numerical and experimental results are set as an objective function to be minimized by making changes to a pre-selected set of parameters in the FE model Since large sets of parameters may lead to ill-conditioning Multi-model approach 1 : multiple sets of relatively few parameters are selected (and independently solved) based on: - direct a-priori knowledge, and/or - extensive simulation identifying most plausible condition states (or damage scenarios) Comparing the resulting multiple solutions enhances reliability 1 Smith et al. 2006

41 Improving FE model-updating (3/3) Results: Model 1 Model 2 Model 3 Model 1 stiffness matrix depending only on the lower columns stiffness I cx,2, I cy,2 Model 2 like 1 + upper columns stiffness: no real (physical) improvement I bx,1 = I bx,2 I by,1 = I by,1 Model 3 like 1 + beams stiffness: reasonably the best physical matching

42 Improving FE model-updating (3/3) Results: Model 1 Model 2 Model 3 HENCE A) Even for simple structures, model-updating is not an easy task, and a multiplemodel approach should be accomplished to depict at which extent results depend on the (arbitrary) choice of the parameters set. B) Small absolute values of f ob are not per se a reliable index of successful updating. C) Observing how the solution improves through enlarging the updating set may provide useful information on its optimal dimension (and on data redundancy). D) Improvement in f ob should always be judged in relative terms: passing from 0.827% to 0.753% may indeed represent a drastic improvement, corresponding to a significantly different solution. E) Since small improvements in f ob may be so important, every care must be taken to minimize all sort of possible errors (in measuring, identification, optimization,...).

43 Conclusions 1) even for simple systems and apparently redundant data, the solution may be extremely sensitive to the choice of the updating parameters as well as to modelling errors 2) testing the whole procedure on a simulated model prior to the real model may provide a helpful insight into such dependence 3) spanning alternative modelling assumptions (multi-model approach) is an effective strategy to increase calibration robustness 4) the influence of unaccounted secondary structural elements (braces) may act as a severe misleading factor in system identification.

44 Risk is the key word of the European Research Project IRIS (Integrated European Industrial Risk Reduction System ) from Integrated European Industrial Risk Reduction System Motivation At present the European Industry recognised their obligation to reconsider risk and safety policies, having a more competitive industry and more risk informed and innovation accepting society in vision. Therefore the large collaborative project IRIS is proposed to identify, quantify and mitigate existing and emerging risks to create societal cost-benefits, to increase industrial safety and to reduce impact on human health and environment. Project Outline The project is led and driven by industry to consolidate and generate knowledge and technologies which enable the integration of new safety concepts related to technical, human, organizational and cultural aspects. The partnership represents over 1 million workers. The proposed project integrates all aspects of industrial safety with some priority on saving human lives prior cost reductions and is particular underpinning relevant EU policies.

45 Objectives Integrated Methodologies for pioneering Risk Assessment and Management New Knowledge-based Safety Concepts Total Safety of Industrial Systems and Networks Knowledge and Technologies for Risk Identification and Reduction Online Monitoring with Decision Support Systems Pattern Recognition in Signal Processing Demonstration & Technology Transfer Standardization & Training Activities Integrated European Industrial Risk Reduction System

46 USA - FHWA Long-term Bridge Performance Program (about 25 MUSD) International Guidelines for the Selection and Management of Technology Applications to Bridges Prepared by United States: A.E. Aktan, F.L Moon, S. Chase, D. Mertz, and N. Gucunski International: H. Wenzel (Austria), Y. Fujino, (Japan), D. Inaudi (Switzerland), J. Brownjohn (U.K.), H. Soon (Korea), and H-Y Koh (Korea) Background and Introduction The objective of the guidelines will be to aid infrastructure owners and practicing bridge engineers in the selection and management of sensor technology applications to bridges. It is stressed that this document is not intended to be a how to guide related to the use of sensors. Rather it will aim to serve as a guide to those who are tasked with the critical responsibilities of (1) identifying the need for sensor technology, (2) ensuring that appropriate approaches are selected, (3) managing the project and ensuring the established best practices are followed throughout the application, and (4) incorporating the results of the application within the decisions-making process. Overview of current bridge engineering and management practice in the US, Europe and the Far East. Summary of Bridge Performance Definitions and Metrics and a discussion of how these vary between the US, Europe and the Far East. Common objectives of infrastructure owners that drive applications of sensor technology to bridges Brief history and description of current practice of technology applications to bridges including brief discussions of proof testing, load testing, NDE, modal analysis, long-term monitoring, etc. Challenges related to employing technology to help inform decisions, inclusive of a wide range of issues such as owner/engineer risk aversion, lack of standards and accountability, cost, liability and indemnification, and the coordination between teams with diverse skill sets, among others. Brief outline of the lessons learned over the last 30 years and a discussion of strategies that may allow for more wide-spread and effective applications of technology in the future. Outline and summary of the report