Upload
others
View
0
Download
0
Embed Size (px)
Citation preview
PERSONAL FITTING PROCEDURE FOR CYCLE ERGOMETER WORKLOAD CONTROL BY ARTIFICIAL NEURAL NETWORKS T. Kiryu et al., IEEE/EMBS October 24, 2002
PERSONAL FITTING PROCEDURE FOR CYCLEERGOMETER WORKLOAD CONTROL BY
ARTIFICIAL NEURAL NETWORKS
T. Kiryu1, K. Shibai1, Y. Hayashi2, and K. Tanaka2
1Graduated School of Science and Technology, Niigata University, 8050 Ikarashi-2, Niigata 950-2181, Japan2Institute of Health and Sport Sciences, University of Tsukuba, 1-1-1 Tennodai, Tsukuba 305-8574, Japan
Abstract-The Internet technology was introduced to provide on-demand support for fitting thecycle ergometer workload personally depending on each individual’s physical work capacity. Itwas useful to combine the ratings of perceived exertion (RPE) with objective physiological indicesduring the exercise. We estimated the RPE from objective indices (i.e., muscular fatigue-relatedindices and the heart rate), using a feed-forward type artificial neural network. Interpreting theestimation errors was useful to design an appropriate workload for an individual elderly person.
Keywords - cycle ergometer, heart rate, muscle activity, ratings of perceived exertion, artificialneural network
PERSONAL FITTING PROCEDURE FOR CYCLE ERGOMETER WORKLOAD CONTROL BY ARTIFICIAL NEURAL NETWORKS T. Kiryu et al., IEEE/EMBS October 24, 2002
Purpose
Health Promotion byExercise
“Wellness”
Social Background- progression of aged society- high motivation for health
Howeverphysical work capacity
- risk for overuse- degeneration of functions- large individual differences
Web-based Healthcare System
IT systemto support Exercise for Wellness
PERSONAL FITTING PROCEDURE FOR CYCLE ERGOMETER WORKLOAD CONTROL BY ARTIFICIAL NEURAL NETWORKS T. Kiryu et al., IEEE/EMBS October 24, 2002
analyzed results
Active 4-bar electrode
Ref. electrode
measured data
control info.
Sports Gymnasium
Network server
PC for maintenance& service
MonitoringIf necessary
Communication portmeasured data
Wellness Bikepaddle sensor
HR sensorInternet
Zoom Up
Home
Health Care Center
���������
amp. network
support center
� ������������
measurement & control unit
�� ���� ��fuzzy parameters
Internet
Browsing by Java made programs
Exercise at any time and at any location
PC for maintenance& service
Continuously Supportingat any time and at anylocation
Internet-based Cycle Ergometer Systemfor Serving Personally Fitted Workload
PERSONAL FITTING PROCEDURE FOR CYCLE ERGOMETER WORKLOAD CONTROL BY ARTIFICIAL NEURAL NETWORKS T. Kiryu et al., IEEE/EMBS October 24, 2002
Estimation of Total fatigue index δδδδ
SEMG, HR
WLWLWL n1n ∆δ ⋅−=+
Recursive Workload Control
WLn : workload at n-th frame
∆WL: incremental step of workload
δ: fatigue indexWorkload control
-Fuzzy Rules-Membership Functions
Surface EMG(SEMG)- amplitude info- frequency-related info.
Heart Rate (HR)
Fuzzy control
Control by Objective Indices
Workload Control for Cycle Ergometer
time
wor
kloa
d
controlled workload
AT
PERSONAL FITTING PROCEDURE FOR CYCLE ERGOMETER WORKLOAD CONTROL BY ARTIFICIAL NEURAL NETWORKS T. Kiryu et al., IEEE/EMBS October 24, 2002
Objective Indices
Amplitude indx
Frequency index
Correlation coefficient betweenARV and MPF
Heart Rate
Surface Myoelectric Signals
m : frame number
emg : samples of ME signals
P(m, f) : power spectrum of ME signal at m-th frame
fH : high-frequency limit for estimation
N : number of samples in a frame
Muscular fatigue index
HR
ARV mN
emg kk m
m N
( ) ( )==
+ −
∑1 1
MPF m
f P m f
P m f
f f
f
f f
fL
H
L
H( )
( , )
( , )
=
⋅=
=
∑
∑
γγγγARV-MPF
fL : low-frequency limit for estimation
1 frame = 5 sec
PERSONAL FITTING PROCEDURE FOR CYCLE ERGOMETER WORKLOAD CONTROL BY ARTIFICIAL NEURAL NETWORKS T. Kiryu et al., IEEE/EMBS October 24, 2002
Segmentation in HR-γγγγARV-MPF Scatter Graph
-1
0
1
60 70 80 90 100 110
seg-1
seg-2
seg-3
µ2-σ2 µ2 µ2+σ2
last
early
µ2-σ2
µ2
µ2+σ2
middle
µ1, µ2, µ3 : mean at each segment σ1, σ2, σ3 : s.d. at each segment
membership functions
heavy
light
γγγγARV-MPFγγγγ
ARV-MPF
Heart Rate
1) Weigh the random values for eachsample,
2) Estimate the center of samples by fuzzymeans, and tentatively segment the samples,
3) Change the weight based on the distancesbetween the center and each sample,
4) Repeat steps 1) - 3) until the centerposition reaches a steady point, and then
5) Divide the samples.
PERSONAL FITTING PROCEDURE FOR CYCLE ERGOMETER WORKLOAD CONTROL BY ARTIFICIAL NEURAL NETWORKS T. Kiryu et al., IEEE/EMBS October 24, 2002
Changes in HR-γγγγARV-MPF Scatter Graph
HR
γARV-MPF
XXXXXXXXXXXXXX
XXXXXXXXXX
XXX
XXXXXX
XXX
X
XXXXXXXXXX
XXXXXXXXXXXXXXXXXXXXX
XXXXXXXXX
XXXXX
XXXXXXXXXXXX
XXX
XXXXXXXXXXXXXXXXX
XXXXX
XXXX
XX
X
XXXXXXXXXXXX-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
60 80 100 120 140 160 180
middle phase
last phase
early phase
Type A
XXXXXXXXXXXXXX
XX
XXXXXXXXXXXXXX
XXXXXXXXXX
XXXXXX
XXXXXXX
XX
XX
X
XX
XXXXXXXX
XXXXXXXX
X
X
X
XXXXXXXXXXXXXXXXX
XXX
XXXXXXXXXXXXXXXXXXXXX
X
XXXXXXXXXX
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
60 80 100 120 140 160 180
γARV-MPF
(a) first set
[bpm]HR [bpm]
(b) fourth set
progressively increasing workload
seven monthsApril, 2000 November, 2000
subject TH (a 70-year-old man)
PERSONAL FITTING PROCEDURE FOR CYCLE ERGOMETER WORKLOAD CONTROL BY ARTIFICIAL NEURAL NETWORKS T. Kiryu et al., IEEE/EMBS October 24, 2002
Objective / Subjective Representation of Fatigue
RPE-Scale
myoelectric signals
electrocardiogram
7 Very, very light 8 9 Very light1011 Fairly light1213 Somewhat light1415 Hard1617 Very hard1819 Very, very hard20
RPE (Ratings of Perceived Exertion)by Borg (Med. Sci. Sports Exerc. 1973)
Objective DataObjective Data
tightSubjective RepresentationSubjective
Representation
PERSONAL FITTING PROCEDURE FOR CYCLE ERGOMETER WORKLOAD CONTROL BY ARTIFICIAL NEURAL NETWORKS T. Kiryu et al., IEEE/EMBS October 24, 2002
Workload Adjustment Procedure
PIWCW1
CW2CW3
PIW
Design of Control Parameters- Fuzzy Rules- Membership Functions
�CW Controlled Workload
�PIW Progressively Increasing Workload
Updating Control Parameters
Key points• classification• segmentation
conventionalapproach
Key point• Personal Fitting
next trial
trial
PERSONAL FITTING PROCEDURE FOR CYCLE ERGOMETER WORKLOAD CONTROL BY ARTIFICIAL NEURAL NETWORKS T. Kiryu et al., IEEE/EMBS October 24, 2002
Evaluation of Personal Fitting Procedure
WL
∆WL
HR
RPE
ANNγARV-MPF
εEstimated RPE
Objective indicesSubjective indices
Estimation errorEstimation
ε( ) ( ) ˆ ( )n RPE n RPE n= −
Estimation of RPEby
Artificial Neural Network
Comparison
PERSONAL FITTING PROCEDURE FOR CYCLE ERGOMETER WORKLOAD CONTROL BY ARTIFICIAL NEURAL NETWORKS T. Kiryu et al., IEEE/EMBS October 24, 2002
Estimation of RPE by ANN
Selection of a suitable ANN
- HR- γARV-MPF
- WL- ∆WL
Estimated RPE
ε( ) ( ) ˆ ( )n RPE n RPE n= −minimum squared error
Numbers of cells at each layer were 4 at the input,4 - 20 at the middle, and 1 at the output (the totalnumber of ANNs checked was 25).
A suitable ANN for personal fitted workload control
PERSONAL FITTING PROCEDURE FOR CYCLE ERGOMETER WORKLOAD CONTROL BY ARTIFICIAL NEURAL NETWORKS T. Kiryu et al., IEEE/EMBS October 24, 2002
Experimental Conditions
18 senior subjects (63.5±7.5 years old)
6 senior subjects (69.3±4.1 years old)
muscles: right leg vastus lateralisgain: 60dBfrequency band width: 5.3Hz - 1.2kHz4-bar active array electrode
LED-photo transistor
lactate in blood every minuterespiration gas exchange every minute
ratings of perceived exertion (RPE)
- Subjects- Subjects
- Myoelectric Signals- Myoelectric Signals
- Heart Rate- Heart Rate
- Metabolism- Metabolism
- Subjective Index- Subjective Index
from November 2001 to December 2001 every 2 weeks
from September 2000 to March 2001 every 2 months
PERSONAL FITTING PROCEDURE FOR CYCLE ERGOMETER WORKLOAD CONTROL BY ARTIFICIAL NEURAL NETWORKS T. Kiryu et al., IEEE/EMBS October 24, 2002
Personal Fitting
temporary increasing workload
Workload Design for Personal Fitting
phase 1phase 2
phase 3phase 4
phase 5phase 6
time
wor
kloa
d
WLmin WLmax
T1 T2 T3
AT
How to determine: - level - timing
Little bit hard exercise andperception of workload changes
Protocol for Exercise
ATmax
ATmin
WL%130WL
WL%70WL
=
=
time
wor
kloa
d
Conventional WL control
AT
AT: Anaerobic Threshold
PERSONAL FITTING PROCEDURE FOR CYCLE ERGOMETER WORKLOAD CONTROL BY ARTIFICIAL NEURAL NETWORKS T. Kiryu et al., IEEE/EMBS October 24, 2002
Sufficient Workload ControlWL
RPE RPE^0
50
100150
510
1520
0 50 100 150 200 250 300 350 400
WL
[W
]
time [frame]
6080
100120
-1-0.500.51
0 50 100 150 200 250 300 350 400
HR
[bp
m]
AR
V-M
PF
time[frame]
HR
ARV-MPFγ γ
-5
0
5
0 50 100 150 200 250 300 350 400time [frame]
ε
Sufficient example for PF
^R
PE, R
PE
Subject KY
2001-Dec-23
-1
-0.5
0
0.5
1
60 80 100 120
AR
V-M
PF
HR [bpm]
γ
33.2% subj.A, Dec 23, 2001
: samples in the middle phase
- HR around AT- No severer muscular fatigue
WLAT= 97 W, WLLT =120 W
Type-A, 73-years-old man
PERSONAL FITTING PROCEDURE FOR CYCLE ERGOMETER WORKLOAD CONTROL BY ARTIFICIAL NEURAL NETWORKS T. Kiryu et al., IEEE/EMBS October 24, 2002
Insufficient Levels and Timing
^
WL
RPE
RPE
^0
50
100
150
5
10
15
20
0 50 100 150 200 250 300 350 400
WL
[W
]
RPE
, RPE
time [frame]
60
80
100
120
-1-0.500.51
0 50 100 150 200 250 300 350 400
HR
[bp
m]
AR
V-M
PF
time[frame]
HR
ARV-MPFγ γ-5
0
5
0 50 100 150 200 250 300 350 400time [frame]
ε
98[W]
106[W]
2001-Nov-11
Subject KY
Insufficient example for PF
6080
100120
-1-0.500.51
0 50 100 150 200 250 300 350 400
HR
[bp
m]
AR
V-M
PF
time[frame]
HR
ARV-MPFγ γ^WL
RPE RPE^0
50100
150
51015
20
0 50 100 150 200 250 300 350 400
WL
[W
]
RPE
, RPE
time [frame]
-5
0
5
0 50 100 150 200 250 300 350 400time [frame]
ε2001-Dec-9
PERSONAL FITTING PROCEDURE FOR CYCLE ERGOMETER WORKLOAD CONTROL BY ARTIFICIAL NEURAL NETWORKS T. Kiryu et al., IEEE/EMBS October 24, 2002
-1
-0.5
0
0.5
1
60 80 100 120
AR
V-M
PF
HR [bpm]
γ
HR-γγγγARV-MPF Scatter Graphs in PF
subj. A
: samples in the middle phase
- HR around AT- No severer muscular fatigue
(b)(a) (c)-1
-0.5
0
0.5
1
60 80 100 120
AR
V-M
PF
HR [bpm]
γ
subj.A, Nov 11, 200120.1%-1
-0.5
0
0.5
1
60 80 100 120
AR
V-M
PF
HR [bpm]
γ
16.6% subj.A, Dec 9, 2001
-1
-0.5
0
0.5
1
60 80 100 120
AR
V-M
PF
HR [bpm]
γ
33.2% subj.A, Dec 23, 2001
Lighter than PIW Relatively light for muscles,but hard for respiration
Appropriate exercise,comfortable
Insufficient timingInsufficient level sufficient workload control
PERSONAL FITTING PROCEDURE FOR CYCLE ERGOMETER WORKLOAD CONTROL BY ARTIFICIAL NEURAL NETWORKS T. Kiryu et al., IEEE/EMBS October 24, 2002
RPE-Workload Scatter Graph atPersonally Fitted Trial
0
2
4
6
8
10
12
14
0 20 40 60 80 100 120
Workload [W]
RPE
WLAT
B B B B
B
B
B
B
BBB
B
B B B
B
B
BB
BB B
B
B
BBB
B
J
J
J
J
J
JJJ
JJJJJJJ
J
J
J
J
JJJJJJJJJJJJ
J
JJJJJJJJJJJJJJ
JJJJJJJJJJJJJJJJJJJJJ
J
JJJJJ
JJJJJJJ
JJJ
JJJJJ
JJJJJ
J
J
JJ
J
JJ
JJ
JJJ
-4-3-2-101234
0 20 40 60 80 100 120
EE
EEEE
EEEEEEEEEEEEEE
EEEEEEEEEEEEEEEEE
EEEEEEEEEEEEEEEEE
EEE
EEEEEEEE
EEEEEEEEEEEEEE
E
E
EE
EEEE
E
E
EE
EE
E
E
EEEEE
E
Workload [W]
RPE estimation error90-194 frame
195-294 frame
WLAT= 97 W, WLLT =120 W
Type-A, 73-years-old manWL
RPE RPE^0
50100
150
51015
20
0 50 100 150 200 250 300 350 400
WL
[W
] ^R
PE, R
PE
ε( ) ( ) ˆ ( )n RPE n RPE n= −
Subject KY2001, Dec. 23
PERSONAL FITTING PROCEDURE FOR CYCLE ERGOMETER WORKLOAD CONTROL BY ARTIFICIAL NEURAL NETWORKS T. Kiryu et al., IEEE/EMBS October 24, 2002
1. The fuzzy rules and membership functions weredesigned based on the individual’s physical workcapacity.
2. Using a feed-forward type artificial neural network, weestimated the RPE (i.e., a subjective index) from theHR and muscular fatigue, and studied the balancebetween the objective and subjective representationsof fatigue.
3. The gap was suitable to determine the appropriatelevels and the timing of temporally inserting workloadsin a basic workload variation.
Conclusions
PERSONAL FITTING PROCEDURE FOR CYCLE ERGOMETER WORKLOAD CONTROL BY ARTIFICIAL NEURAL NETWORKS T. Kiryu et al., IEEE/EMBS October 24, 2002
Conclusions (continued)
1. Evaluation of basic data under progressivelyincreasing workload.
2. Continuously changing the intensity and timingof temporary increasing workloads.
3. Comparison between subjective index (RPE)and objective indices by ANN.
- small error between RPE and estimated RPE.
- RPE greater than 13 and no progression of muscular fatigue.
Personally Fitting Procedure
Wellness Bike for Home-Based Exercise