How to avoid statistical outliers under special speed (acceleration and deceleration)

Hello dear community,

I use and plot some measure data (rotating speed of 4 motors). The measures work great up to speed of 30-40 1/s. Under this speed the measured values sometime get some statistical outliers (look screenshot) with high values. I'm already used a median filter, but the filter changes also some "good" data values and passed some outliers.

Do you have an idea how can I avoid this values and set the values in right range?

Maybe interpolaton?

Or using of median Filter just under 30-40 Hz?

Example values (outlines on beginning):

0

0

0

0

0

0

35.9221208420145

10.8085906678628

15.4286063840078

18.5873605947955

20.6104824913951

22.2443681126636

23.6116358141292

24.6560481286060

25.5008374435171

26.1552791927909

26.7056931426144

27.1928800542243

27.5813385525270

27.8667967117180

28.0977802753583

28.2559302961520

28.3540485870989

28.5013965684319

29.0721344386347

31.4844461301742

37.3895684126500

39.7713149391002

41.3880651693887

43.7340097526842

45.0987680637302

45.1770937663817

45.5087891515580

45.0346762609292

44.6965102588530

44.2460477779540

44.8309871783377

45.2556960082974

45.4724575012369

45.6825445034530

47.5853103775104

48.0169019494862

48.7487797333042

49.1463273343138

50.6549677435695

Example values (outlines on the end):

92.7870315625418

91.8920602119135

91.9215499840484

91.6945182796246

90.8861434488664

90.0099028473191

91.8584443431570

91.3844784177349

90.0810712046883

90.9688215024709

93.1290412001423

93.3871501105416

95.3125315594778

94.0114693992667

96.2167553022808

94.8037543768042

93.0367194307825

92.7335897091072

91.5667063455728

91.2529014457080

91.3373726929620

90.2445627650934

90.6590942713811

89.9262569565512

89.0005722458669

89.3424413764390

87.9258263224508

87.4368676356960

88.4529586175770

89.1186168790660

85.4503219449361

85.8958941762584

86.3707030575229

83.2867895195613

80.0216058335751

76.1143109101614

74.4114068617345

73.9287728040303

70.0224085383574

74.1583014678347

72.9980290532156

67.7454333028347

58.9773353095348

51.6902718908302

45.9760701738506

41.3103647705209

37.2113792397715

33.8588979365184

30.9228246623297

28.3032740033323

26.2249671931807

24.1875396000099

22.2651696471118

20.5769784764805

19.2888836581187

18.0384941465087

16.7995702984910

15.7064773512597

14.6524440276638

13.5649621267317

12.6668862260279

11.7483046875093

10.8316544268972

10.0875600209821

9.32062000764291

8.71383757406762

8.09775611178143

348.432055749129

59.0702345088310

28.5013965684319

19.8515106999643

13.8392980708019

9.81836033382425

0

43.7024735600035

0

8.35757028716612

0

0

0

0

0

0

0

0

0

0

Thank you!

Answer 1

[Steven Lord]

Use isoutlier to detect the outliers, rmoutliers to remove them, and/or filloutliers to replace them.

[Nik Rocky]

Great, thank you!

I'm using:

v1 = filloutliers(v1,'previous','movmean',30);
v2 = filloutliers(v2,'previous','movmean',30);
v3 = filloutliers(v3,'previous','movmean',30);
v4 = filloutliers(v4,'previous','movmean',30);

and get:

its allready better, but not away. I follow working to solve it!

[Nik Rocky]

here is zoomed view of "problem values"

[Steven Lord]

If you have time-based data where the data may be irregularly sampled, you might want to specify duration data as the 'SamplePoints' option (the number of seconds since the start of the data collection when each data point was sampled) and specify the window length as a duration array as well. See the "Determine Outliers with Sliding Window" example on the documentation page for filloutliers for an illustration of this technique.

[Nik Rocky]

Thank you Steven!