A neural network? I think probably not. All you need to do is characterize the system under load. Of course if you goal is to move a motor N number of degrees, the only way you are going to know if the motor has moved that is with a rotor encoder on the output shaft of the motor. Without that you are blind the the motors position, velocity, and accelerations leaving you to only able guess. The only way commercial robots, like the ones used in factories, know where they are is by sensing their the speed and position of the motors. Also make sure that your motors have enough torque for your load. Trust me, if you have mechanical issues, there is nothing you can do on the control side to compensate for that. On Mon, Mar 9, 2015 at 9:32 AM, Tobias Gogolin wrote: > Thanks for all the replies so far! > It is like I expected - if anybody is doing it they are keeping it > their proprietary secret... > > between dynamic load of having to achieve the next motion target and > the associated acceleration effort to guesstimate how much mechanical > resistance is present might be something a neuronal network might > learn, because the simple current measurement (I suspect) has poor > signal to noise ratio! I wonder how a 'lost step' would look and if > the voltage measurement of the phases while not driven might provide > hints if and before it happens? > A little more background is that I am speculating if and when we get > our robots to 'feel' more? I guess for now the force transducer or > load cell.will be the solution meanwhile some wiser engineers do their > dissertation about how much can be deducted from the motor wires... > > cheers > > Tobias > > On 3/9/15, Robert Rolf wrote: >> Given the context of the question, robotics, OP seems to be looking >> for "accurate effort diagnosis", e.g.. force/torque required. >> >> Given that the robot geometry and masses are known, the required >> torques for a given motion/position can be computed (or measured by movi= ng >> slowly through the range of motion envelop and recording 'static' curren= ts >> as a starting point). >> >> Measuring the actual currents on your BLDC motors will give your torques >> pretty >> much directly. >> >> Not sure how you can measure force/torque output of a stepping motor >> from the input side. Perhaps with a precision position sensor, and seein= g >> how >> far the commanded position deviates from 'expected' because of torque >> pulling the pole off step center, and knowing what the coil current is >> at that exact >> moment. >> >> Since most motors are geared in robotics applications, the only 'good' >> measure >> of 'effort' is with a force transducer. e.g. load cell. >> >> Robert >> >> On Sun, Mar 8, 2015 at 4:57 PM, IVP wrote: >>>> Would you need designated hardware, or could it be done all >>>> electronically via given wires (Stepper or bldc servo)? >>> >>> A driver like the L298 has low-side sense resistor in series with >>> the motor coil. The voltage across the resistor is passed back to >>> eg an L297, which is commonly used with the L298. The L297's >>> comparator uses the coil voltage vs a Vref to determine the coil >>> current, for chopping etc. Vsense is available to any IC >>> >>> http://www.uni-kl.de/elektronik-lager/419182 >>> >>> As Jason shows, a discrete circuit can be used to deliver Vsense >>> to an ADC >>> >>> I'm not sure what you mean by "Resistance" in the Subject. Do >>> you mean the amount of current needed as the motor tries to >>> move against a load ? If the coil is being chopped or PWMed, >>> then the start of a drive cycle can be used as the start point for >>> gathering data to track the increasing coil current >>> >>> Joe >>> >>> >>> ----- >>> No virus found in this message. >>> Checked by AVG - www.avg.com >>> Version: 2015.0.5751 / Virus Database: 4299/9254 - Release Date: 03/08/= 15 >>> >>> -- >>> http://www.piclist.com/techref/piclist PIC/SX FAQ & list archive >>> View/change your membership options at >>> http://mailman.mit.edu/mailman/listinfo/piclist >>> >> -- >> http://www.piclist.com/techref/piclist PIC/SX FAQ & list archive >> View/change your membership options at >> http://mailman.mit.edu/mailman/listinfo/piclist >> > > > -- > Tobias Gogolin > Tel. D1 (49) 0151 5187 5210 > Tel. D2 (49) 0152 0839 5060 > skype: moontogo > messenger: usertogo@hotmail.com > > You develop Sustainable Ranch Technology at > http://tech.groups.yahoo.com/group/SURA-TECH > an Open Source Electric Motor/Alternator at > http://groups.yahoo.com/group/Performance_Axial_Flux > and an Open Source Motor Controller at http://groups.yahoo.com/group/GoBo= x > -- > http://www.piclist.com/techref/piclist PIC/SX FAQ & list archive > View/change your membership options at > http://mailman.mit.edu/mailman/listinfo/piclist --=20 Jason White --=20 http://www.piclist.com/techref/piclist PIC/SX FAQ & list archive View/change your membership options at http://mailman.mit.edu/mailman/listinfo/piclist .