Adaptive environment perception in Cyber-Physical Systems Sebastian - PowerPoint PPT Presentation

Adaptive environment perception in Cyber-Physical Systems Sebastian Zug, Andr´ e Dietrich, Christoph Steup and J¨ org Kaiser Dept. of Embedded Smart Systems (ESS) Institute of Distributed Systems University Magdeburg 7th Workshop on Adaptive and Reconfigurable Embedded Systems (APRES 2015) April 13, 2015, Seattle, USA

Introduction and motivation Concept Evaluation Conclusion Motivation Focus of the presentation Introduction and 1 motivation Juxtapose Traditional Systems vs. CPS Challenges Known environments on Concept 2 run-time Adaptive Sensing Statically configured Controller sensor actuators systems Sensing optimization Optimized adjustment Evaluation 3 between sensors and Conclusion 4 application

Introduction and motivation Concept Evaluation Conclusion From static controlled systems to CPS Known environments on Occurrence of run-time unexpected situations Static configured sensor Adaptive cooperation of actuators systems smart X Optimized adjustment Online-adjustment of between sensors and sensors and applications application needed

Introduction and motivation Concept Evaluation Conclusion Challenges of adaptive sensor aggregation Interpretation of unknown data sets Indentification of relevant sensors Transformation, filtering, evaluation, synchronisation of data sets Critical temporal adjustment o 1 Sensor 0 Sensor 1 o 0 Delay uniform Rayleigh 0 p 1 Period 90ms 60ms 2 2 p p o p Offset 35ms 18ms 1 p 0 application 1 measurment count 2 2 App period p p = 45 ms 0 Super period p super = 360 ms S e 0 n s o r 1 S e o r n s

Introduction and motivation Concept Evaluation Conclusion Challenges of adaptive sensor aggregation Interpretation of unknown data sets Indentification of relevant sensors Transformation, filtering, evaluation, synchronisation of data sets Critical temporal adjustment Optimized Schedule o 1 o 1 o 0 o 0 0 1 p 1 p 1 2 2 2 1 p p p p o p o p 1 1 p 0 p 0 application application 1 1 measurment measurment count count 2 1 2 2 0 1 S e 0 S e 0 n s o r n s o r 1 1 S e o r S e o r n s n s

Introduction and motivation Concept Evaluation Conclusion Evaluation criteria of a sensor set Which goals of an application should be monitored? High-level Low-level Precision Minimize variance of the input count Accuracy Maximize minimum input count Reliability Minimize maximal age Minimize average age Minimize mean input uncertainty Minimize failure probability

Introduction and motivation Concept Evaluation Conclusion Adaptive Sensing Controller (ASC) start run Two level sensing new sensors new sensors available available evaluation Data sheet Application Error state evaluation D not no suit- A based on static relevant able sensor application relevant adjusted new sensors data sheet Static Configu- information analysis ration no need for no valid result C B optimization specific analysis considering necessary processing dynamic not Network offset completed monitoring detemined E parameters completed (offsets, delays) Optimi- Dynamic additionally Analysis zation valid no valid result G F schedule(s) found

Introduction and motivation Concept Evaluation Conclusion Mathematical model - Static analysis I 1. Individual analysis of sensor n p s < p p � � p p count of m max + 1 p s measurements � � p p m min p s � � p p p p count P ( m = m max ) p s − p s probability � � 1 − p p p p P ( m = m min ) p s + p s 2. Calculation of the multi sensor result Convolution Sensor 0 Sensor 1 1 1 1 k = ∞ = * P n ( m ) � = P ( k ) P ( m − k ) 0 0 0 0 2 4 6 8 0 2 4 6 8 k = −∞ count 0 2 4 6 8

Introduction and motivation Concept Evaluation Conclusion Mathematical model - Static analysis II 2. Calculation of the multi sensor result Sensor 0 Sensor 1 Convolution 1 1 1 k = ∞ = * P n ( m ) � = P ( k ) P ( m − k ) 0 0 0 k = −∞ count 0 2 4 6 8 0 2 4 6 8 0 2 4 6 8 3. Comparism to application demands at least 6 measurement values per cycle variance of measurement count smaller than x no measurements older than x ms ... A more detailed analysis is necessary!

Introduction and motivation Concept Evaluation Conclusion Mathematical model - Dynamic Analysis 1. Individual analysis of sensor n Offset p o Delay related to communication Example: The maximum/minimum age can now calculated by a max = p s − mod( o s , gcd( p s , p p )) a min = gcd( p s , p p ) − mod( o s , gcd( p s , p p )) 2. Optimize o p Related to demands of the application Maximum age of the measurement set Variance of the age Uncertainty ...

Introduction and motivation Concept Evaluation Conclusion Implementation Goals Implementation of the ASC as a web-service Evaluation platform for static configured applications

Introduction and motivation Concept Evaluation Conclusion Example - Model car tracking Parameters Car speed approx. 3.2m/s Control algorithm scheduled each p s = 60 ms 2 cameras observing with a sensor period p s = 40 ms Position measurements are disturbed by Gaussian noise σ s = 2 cm .

Introduction and motivation Concept Evaluation Conclusion Example - Fusion and estimation approach 1. Synchronization of individual sensors by estimating the position at fusion time ˆ x n ( k · t p ) = x n ( t ) + f ( k · t p − t ) Core idea ˆ σ n ( k · t p ) = σ s + σ v · ( k · t p − t ) The age of a data set is mapped on the 2. Fusion of all position estimation uncertainty of the position estimation. i i ˆ x n � � σ n 2 · ˆ x = ˆ σ n 2 ˆ n = i n = i 1 σ 2 = ˆ � i 1 n = i σ n 2 ˆ

Introduction and motivation Concept Evaluation Conclusion Example - Results of the optimization 20 3 . 5 ∆ o between cameras 3 10 2 . 5 2 0 0 10 20 30 40 50 0 1 2 Fusion offset o p improvement of ˆ σ Optimization goal Chose o p in a way, that minimizes the resulting uncertainty ˆ σ n

Introduction and motivation Concept Evaluation Conclusion Conclusion and future work Next steps Extended mathematical model considering communication delays Real-World implementation of the car-tracking example Website for testing purposes Future steps Implementation in a more realistic heterogeneous scenario Integration of a network monitoring tool in order to determine the specific data dynamically Development of a concept for configuration switches

Introduction and motivation Concept Evaluation Conclusion start run new sensors new sensors available available Data sheet Application Error state evaluation no suit- D not A relevant able sensor application relevant adjusted new sensors Static Configu- Thanks for your ration analysis no need for no valid result C B optimization interest! specific analysis necessary processing not Network offset completed monitoring detemined E completed Dynamic Optimi- Analysis zation valid no valid result G F schedule(s) found