Difference between revisions of "ICub joint sensor docs"

From Wiki for iCub and Friends
Jump to: navigation, search
m (Hardware description:)
m (Sensitivity)
 
(27 intermediate revisions by 2 users not shown)
Line 1: Line 1:
= Hardware description: =
+
= Hardware description and testing setup =
 +
 
 +
Since the tests include measures in closed loop it is better to experiment directly on the BLL that will do the control.
  
 
The joint is controlled by a BLL board.
 
The joint is controlled by a BLL board.
Line 5: Line 7:
 
* a standard AEA absolute encoder (4096 counts for revolution) placed after the spring (used to have absolute calibration)
 
* a standard AEA absolute encoder (4096 counts for revolution) placed after the spring (used to have absolute calibration)
 
* an AVAGO optical encoder (80000 counts for revolution) placed after the spring, providing fine position information.
 
* an AVAGO optical encoder (80000 counts for revolution) placed after the spring, providing fine position information.
* a XXXX (code name missing) magnetic linear sensor, measuring the deflection of the spring (and thus the torque T=k*x).
+
* a XXXX (code name missing, based on AS5311) magnetic linear sensor, measuring the deflection of the spring (and thus the torque, computed as T=k*x).
 +
* a XXXX (code name missing) reflective optical encoder, (XXXX counts for revolution) mounted on the motorside (=before the spring and the gearbox)
 +
The joint is also equipped with a 6 axis F/T sensor at the end-effector.
 +
* The weight of the end-effector (after the F/T sensor) is XXXX N.
 +
* The total weight (end-effector+F/T sensor+support) sensed by the spring is XXXX N.
 +
 
 +
= Testing =
 +
 
 +
The following tests have been proposed for the sensor:
  
= Test =
 
Proposta di test per il sensore:
 
 
=== Sensitivity ===
 
=== Sensitivity ===
 
1. sensibilità  minima e ripetibilità (applicando e rimuovendo lo stesso carico sul giunto, quanto si discosta la misura del sensore da quella stimata dal modello).
 
1. sensibilità  minima e ripetibilità (applicando e rimuovendo lo stesso carico sul giunto, quanto si discosta la misura del sensore da quella stimata dal modello).
 +
Applicare un peso calibrato al sensore verificare che non ci sia isteresi.
 +
 +
''First tests using the soft spring 180Nm/rad:''
 +
 +
[[Image: KneeCalib1.jpg|300px]]
 +
 +
From the above plot, it can be observed a large hysteresis which has to be investigated. The hysteresis decreases with the increase of the applied load.
 +
 +
''First tests using the hard spring 350Nm/rad:''
 +
 +
[[Image: KneeCalib2.jpg|300px]]
 +
 +
Also in this plot the hysteresis is present, but it's constant with the increase of the applied load. The amount of the hysteresis is approximately comparable with the smallest one observed @ 10Nm with the soft spring.
 +
 +
By removing the offset, it's possible to compare the two plots:
 +
 +
[[Image: KneeCalib3.jpg|300px]]
 +
 +
From the above plot, it can be noticed a non-linear response of the soft srping (plus the previously observed hysteresis).
 +
 +
* The maximum deflection of the spring is +- 15 degrees (corresponding to about +-40Nm) The estimated spring constant is thus XXXX Nm/deg. Please check this data .
 +
* The accuracy of the defection sensor is 4096 counts for linear magnetic pole (2mm) This corresponds to a theoretic resolution of 0.488um = XXXX Nm. Please check this data .
 +
 +
[[Image: FT1.jpg|300px]]
 +
[[Image: FT4.jpg|300px]]
 +
[[Image: FT2.jpg|300px]]
 +
[[Image: FT3.jpg|300px]]
  
[[Image: FT1.jpg|200px]]
+
* The tests performed by Alberto show a very good linearity of the spring can be observed (deflections sensor vs AVAGO encoder).
[[Image: FT2.jpg|200px]]
+
* The 'wavy' pattern probably corresponds to the change of magnetic pole on the strip. The amplitude of the waves is very small (about 10^-2 degs corresponding to XXXX Nm) Please check this data.
[[Image: FT3.jpg|200px]]
+
* A strange effect can be also observed. When the joint reaches its hardware limit, an offset is introduced in the deflection measurement (plot n2). Further investigation is required.
[[Image: FT4.jpg|200px]]
 
  
 
=== Noise immunity ===
 
=== Noise immunity ===
 
2. Immunità della misura al rumore elettromagnetico (i.e. generato dalla PWM del motore brushless associato). Giunto controllato in posizione.
 
2. Immunità della misura al rumore elettromagnetico (i.e. generato dalla PWM del motore brushless associato). Giunto controllato in posizione.
  
=== Drift ==
+
=== Drift ===
 
3. Drift in posizione statica a lungo termine (=8ore). Si spera che non ce ne sia proprio! ). Giunto controllato in posizione.
 
3. Drift in posizione statica a lungo termine (=8ore). Si spera che non ce ne sia proprio! ). Giunto controllato in posizione.
  
4. Dirft dopo 8 ore di cicli a velocità random variabile (per simulare il normale funzionamento). Anche qui si spera drift = 0. Giunto controllato in posizione.
+
4. Dirft dopo 8 ore di cicli a velocità random variabile (per simulare il normale funzionamento). Anche qui si spera drift = 0. Giunto controllato in posizione. Il limite hardware non deve essere raggiunto. Provare anche con carichi (pesi) diversi applicati all'end-effector.
  
 
=== Force Control, system identification ===
 
=== Force Control, system identification ===
5. Identificazione sistema in ciclo chiuso, ingresso chirp. Giunto controllato in coppia.
+
5. Identificazione sistema in ciclo chiuso, ingresso chirp. E' preferibile bloccare l'estremita' del giunto. Giunto controllato in coppia (PWM=Kp*(T-Td)).
  
 
=== Calibration ===
 
=== Calibration ===

Latest revision as of 15:05, 1 March 2013

Hardware description and testing setup

Since the tests include measures in closed loop it is better to experiment directly on the BLL that will do the control.

The joint is controlled by a BLL board. The joint is equipped with 4 position sensors:

  • a standard AEA absolute encoder (4096 counts for revolution) placed after the spring (used to have absolute calibration)
  • an AVAGO optical encoder (80000 counts for revolution) placed after the spring, providing fine position information.
  • a XXXX (code name missing, based on AS5311) magnetic linear sensor, measuring the deflection of the spring (and thus the torque, computed as T=k*x).
  • a XXXX (code name missing) reflective optical encoder, (XXXX counts for revolution) mounted on the motorside (=before the spring and the gearbox)

The joint is also equipped with a 6 axis F/T sensor at the end-effector.

  • The weight of the end-effector (after the F/T sensor) is XXXX N.
  • The total weight (end-effector+F/T sensor+support) sensed by the spring is XXXX N.

Testing

The following tests have been proposed for the sensor:

Sensitivity

1. sensibilità minima e ripetibilità (applicando e rimuovendo lo stesso carico sul giunto, quanto si discosta la misura del sensore da quella stimata dal modello). Applicare un peso calibrato al sensore verificare che non ci sia isteresi.

First tests using the soft spring 180Nm/rad:

KneeCalib1.jpg

From the above plot, it can be observed a large hysteresis which has to be investigated. The hysteresis decreases with the increase of the applied load.

First tests using the hard spring 350Nm/rad:

KneeCalib2.jpg

Also in this plot the hysteresis is present, but it's constant with the increase of the applied load. The amount of the hysteresis is approximately comparable with the smallest one observed @ 10Nm with the soft spring.

By removing the offset, it's possible to compare the two plots:

KneeCalib3.jpg

From the above plot, it can be noticed a non-linear response of the soft srping (plus the previously observed hysteresis).

  • The maximum deflection of the spring is +- 15 degrees (corresponding to about +-40Nm) The estimated spring constant is thus XXXX Nm/deg. Please check this data .
  • The accuracy of the defection sensor is 4096 counts for linear magnetic pole (2mm) This corresponds to a theoretic resolution of 0.488um = XXXX Nm. Please check this data .

FT1.jpg FT4.jpg FT2.jpg FT3.jpg

  • The tests performed by Alberto show a very good linearity of the spring can be observed (deflections sensor vs AVAGO encoder).
  • The 'wavy' pattern probably corresponds to the change of magnetic pole on the strip. The amplitude of the waves is very small (about 10^-2 degs corresponding to XXXX Nm) Please check this data.
  • A strange effect can be also observed. When the joint reaches its hardware limit, an offset is introduced in the deflection measurement (plot n2). Further investigation is required.

Noise immunity

2. Immunità della misura al rumore elettromagnetico (i.e. generato dalla PWM del motore brushless associato). Giunto controllato in posizione.

Drift

3. Drift in posizione statica a lungo termine (=8ore). Si spera che non ce ne sia proprio! ). Giunto controllato in posizione.

4. Dirft dopo 8 ore di cicli a velocità random variabile (per simulare il normale funzionamento). Anche qui si spera drift = 0. Giunto controllato in posizione. Il limite hardware non deve essere raggiunto. Provare anche con carichi (pesi) diversi applicati all'end-effector.

Force Control, system identification

5. Identificazione sistema in ciclo chiuso, ingresso chirp. E' preferibile bloccare l'estremita' del giunto. Giunto controllato in coppia (PWM=Kp*(T-Td)).

Calibration

6. [calibrazione] Verificare che tra giunti diversi la k (gain di calibrazione) sia uguale e ripetibile (richiede più giunti su cui provare).

7. [calibrazione] Verificare che la tara (offset di calibrazione) sia mantenuta nel tempo (simile al punto 1, ma mi era sembrato di capire dai grafici e da una discussione con Alberto che se il giunto urta il limite hardware, rimane un bias. Da verificare).

Notes, comments

None.