Short-Transient Discrete Time-Variant Filter Dedicated for Correction of the Dynamic Response of Force/Torque Sensors
Round 1
Reviewer 1 Report
English prose a bit rough in places, but overall the paper is well written and well supported and experimentally validated.
Author Response
Dear Reviewer,
We would like to thank you for your comment regarding our paper. Your suggestion has been incorporated in the revised manuscript. The English has been improved in the paper.
Sincerely yours,
Authors
Reviewer 2 Report
I put in the point of view of the ignorant person who may read your paper because it is interested in your work, so I have a few comments
abstract: "it has been proposed" means by you (in this case I'd put "our proposed approach is..." or something similar) or it's a widespread investigation from several groups? (in this case it's ok but maybe a few citation needed?)
line 188: there seems to be a little argumentative sentence, I am aware that experimental activity often relies on intuition and experience in driving the decisions rather than rigorous mathematics, can you please rephrase this a bit?
table 1, it seems that the time-varying filtered signal never goes beyond 1% of error, so it is not clear why you quote the settling time by the 2% criterion: there is a typo/error in reported values?
line 196, what do you mean with the term horizon? the range of allowed variation of parameters or the time window during which the parameters vary from the transient-optimized values to the stationary-optimized ones? also, the effect is not clear to me of parameters varying on a timescale longer than the settling time in case the second option is valid.
figure 9 and 10, which is the meaning of the y axis? I guess the x-axis is teh sample number from the transient, but this is not explained
again, it is not clear to me if the filter is initialized to the optimal parameters for the transient case and then it is "triggered" to change the parameters in the way you studied once a transient occurs, or if this is just an offline, non causal (i.e. externally triggered by yout test setup) optimization: in this case it would be very great to have future validations on real world signals, that may be noisy and so on, with a robust trigger which allow the user to effectively start the parameter sweeping.
again, it does not make much sense to me the "k" parameter as it seems not related to a physical quantity, differently from the natural frequency and damping parameters.
also, a few typos are still present but the paper is well written
Author Response
Dear Reviewer,
We would like to thank you for your comments regarding our paper. Your suggestions have been incorporated in the revised manuscript.
abstract: "it has been proposed" means by you (in this case I'd put "our proposed approach is..." or something similar) or it's a widespread investigation from several groups? (in this case it's ok but maybe a few citation needed?)
Answer: Done.
1.***************************
line 188: there seems to be a little argumentative sentence, I am aware that experimental activity often relies on intuition and experience in driving the decisions rather than rigorous mathematics, can you please rephrase this a bit?
Answer: The sentence has been reformulated.
The selection of functions was based on two rules. The first is: the larger the undamped natural frequency is, the shorter transient of the filter. The second is: the larger value of the damping factor, the smaller overshoot of the step response.
2.***************************
table 1, it seems that the time-varying filtered signal never goes beyond 1% of error, so it is not clear why you quote the settling time by the 2% criterion: there is a typo/error in reported values?
Answer: We chose the 2% settling time measure as it is commonly used in the control engineering. Moreover, the 2% settling time is often a default value in signal processing tools.
3.***************************
line 196, what do you mean with the term horizon? the range of allowed variation of parameters or the time window during which the parameters vary from the transient-optimized values to the stationary-optimized ones? also, the effect is not clear to me of parameters varying on a timescale longer than the settling time in case the second option is valid.
Answer: By 'horizon' we mean the time span of the possible parameters variation. If the parameters would vary further in time and those changes would by as dynamic as at the beginning, it would deteriorate the amplitude behaviour, i.e. the steady state would be achived later, which actually is the opposite to our goals.
4.***************************
figure 9 and 10, which is the meaning of the y axis? I guess the x-axis is teh sample number from the transient, but this is not explained
again, it is not clear to me if the filter is initialized to the optimal parameters for the transient case and then it is "triggered" to change the parameters in the way you studied once a transient occurs, or if this is just an offline, non causal (i.e. externally triggered by yout test setup) optimization: in this case it would be very great to have future validations on real world signals, that may be noisy and so on, with a robust trigger which allow the user to effectively start the parameter sweeping.
again, it does not make much sense to me the "k" parameter as it seems not related to a physical quantity, differently from the natural frequency and damping parameters.
Answer: You are right with the X-axis. As for the Y-axes values, they are the actual values of the particular parameters, as stated in each title.
The optimal set of parameters were designed based on the identification of the filters in the sensor. In the experimental setup only the RAW signals (filtered and unfiltered) were acquired along with the precise moment of the sensor excitation. After that the experiments with the time varying filters were conducted based on that input signals (unfiltered sensor output and excitation moment) in the offline state and compared with the classical stationary systems (filtered senseor output). Of course the real-life validation of the time-varying systems with the mentioned sensor is planned.
As for the physical representation of the "k" parameter You are right, but its influence on the overall beaviour of the systems transient response is also siginificant. Therefore, we decided to incorporate it to our time-varying model.
Sincerely yours,
Authors
Reviewer 3 Report
This paper describes the application of a discrete time-variant filter to reduce the transient time in the response of force/torque sensors, which is critical for robots interacting with their environment.
The proposal itself, as well as the simulations and analysis done so far seem promising, however, the novelty and significance of this work are not explained nor highlighted in the abstract and introductory sections, and the lack of justification of certain statements makes this issue much less clear.
In this sense, I find the state-of-the-art somehow incomplete. Only a couple of papers are said to work on improving the dynamic properties of force sensors while robots establish contact with the environment [11,12], and only one on time-variant filtration for JR3 sensors [15]. Are not any other proposals? What about other force sensors? As deduced from what has been written, it seems that nobody has followed this research line in 4 years.
In Section 2 the authors identify the type of filter included in the commercial sensor basing on “some additional test and investigations” that are not provided/explained with specific data nor referenced to previous work.
In Section 3, data about dynamical parameters of the elliptic filters should not appear in the text, but in tables. As well as in Section 4 about the piezoactuator parameters. On the contrary, in Section 5 table captions are too long and that text should be moved to the corresponding paragraph for further elaboration.
Continuing in Section 5, it is said that in a first step the way parameters vary with time is establish according to authors’ previous experience and won’t be discussed in the paper. In that case, if authors are not going to include a justification here, they should, at least, provide the functions ai(w) and bi(w) and include the corresponding reference.
This tendency appears again in Section 6, where previous research is cited but no specific reference comes with that statement.
As previously said, this repeated lack of justification of assertions included in the text hinders to narrow down which proposals are original contributions that give rise to significant advances in the area.
Coming back to Section 4, I suggest modifying Fig 2 so that sensor position and fixation method could be seen. With respect to Fig. 3, how was the settling time measured? According to the definition provided in tables1 and 2? If so, please consider moving the definition to Section 4.
Has the gun-based method been used before? Have the measurements been taken several times to ensure the repeatability of the experiment and the control of the impulse generation?
Moving forward to Section 5, I suggest providing data about the HIL testing platform used. For example, about real-time processing and I/O options.
In Section 6 authors establish that “In most test cases it allowed additional 10% improvement of the settling time in comparison to first stage optimization alone.” Please, elaborate more about the comparative improvement in all the cases and provide measurements and values. Besides, authors claim the superior performance for an impulse stimulus, what about the step response?
Regarding Fig. 9 and 10, further explanation is needed. Graphs are so small that it is difficult to extract conclusions. For example, does wo remain constant in Fig.9? and k in Fig 10? Either way, graphs should never be self-explanatory. Please, elaborate a bit more about parameter variations and explain titles (which is/are the current coefficients parameter?)
Other minor aspects:
Please, put references in order of appearance: in line 90 references are not consecutive and, thus, references 20, 21, 23 and 24 appear in the text later than 22 and 25.
English needs revision, there are some articles missing and prepositions mistaken, as well as a couple of inconsistences between the plural and the singular forms of nouns/verbs.
Please, for better understanding, please rephrase/correct some sentences:
Line 26: “Current measurement in the cascade robot control structure is one of key method used as indirect determination of forces and torques acting at a kinematic chain”
Line 29: “Precise current measurements paved the way to force interactions with the environment,…”
Line 36: “One of the approaches to a solution to this problem includes introduction of artificial compliance on the site of the robot or on the site of the environment.”
Line 38: “Still, the most attractive approach includes lack of any required changes of the robot structure.”
Line 51: “Based on JR3 sensor experimental environment has been prepared for verification purposes.”
Line 55: “Properties of commercial filters implemented” What has been implemented? The properties or the commercial filters?
Line 66: “A reduction in duration of transients in the filters applied would enable one to decrease the strength of the impulse-like forces.”
Line 98: “Parameter h0 is calculated such that H(jw)max = 1.”
Line 118: “Of course one can not use the transfer function representation for the time-varying systems.”
Line 125: “The JR3 torque/force sensors clearly contributed to the development of commercial robotic systems interact with the environment.”
Line 152: “Despite all measures, used, a special experiment was prepared in order to determine the impulse duration.”
Line 169: “In this paper we investigated the possibility of designing and performance of the 500Hz and 125Hz filters (first and second in the cascade of hardware filters) as a digital time-varying structures.”
Lines not provided (page 9): “We introduced the Bézier like curves as the base of the base of such non linearity.”
Line 209: “The last point is given by the final, predetermined by the time-invariant design value. In this paper we run the random placement of the one middle control point.”
Author Response
Dear Reviewer,
We would like to thank you for your comments regarding our paper. Your suggestions have been incorporated in the revised manuscript.
1.***********************
The proposal itself, as well as the simulations and analysis done so far seem promising, however, the novelty and significance of this work are not explained nor highlighted in the abstract and introductory sections, and the lack of justification of certain statements makes this issue much less clear.
Answer: The main novelty of the paper is the proposal of the discrete time, optimized, time-varying filter that can be used to reduce the transient state of the force/torque sensor response.
Continuous time-variant filtration in previous research, proved to improve dynamic properties of the sensor in terms of reducing the duration of transient time. Unfortunatelly, the use of continuous time domain in the time-varying filters lead to a significant limitations in terms of their practical usage. This article presents a discrete form of time-variant filters, opening a new way to their implementation. Moreover the proposed optimization routine, further improved the dynamic properties of the filter in terms of its transient response.
2.*********************
In this sense, I find the state-of-the-art somehow incomplete. Only a couple of papers are said to work on improving the dynamic properties of force sensors while robots establish contact with the environment [11,12], and only one on time-variant filtration for JR3 sensors [15]. Are not any other proposals? What about other force sensors? As deduced from what has been written, it seems that nobody has followed this research line in 4 years.
Answer: First of all, there are only few companies in the world that offer 3D force/torque sensors for industrial robots. A similar company to JR3 is for example ATI. According to their manual, the sensor has also built-in classical low-pass filter with a predefined cutoff frequency 5Hz to 3000Hz. In the literature we have not found methods to improve dynamic properties of force/torque sensors. The only exception is the use of Kalman filtration. However, this approach is so different from the proposed method that we did not point out these works in the paper. It should also be noted that the JR3 force sensor used in the experiment costs 10k Euro. Unfortunately, we did not have enough budget to test other types of sensors.
3.********************
In Section 2 the authors identify the type of filter included in the commercial sensor basing on “some additional test and investigations” that are not provided/explained with specific data nor referenced to previous work.
Answer: In the technical note of the JR3 sensor, one can find the frequency response plots of the implemented filters. We identified the filters based on the passband peak-to-peak ripple and minimum stopband attenuation.
4.*******************
In Section 3, data about dynamical parameters of the elliptic filters should not appear in the text, but in tables. As well as in Section 4 about the piezoactuator parameters. On the contrary, in Section 5 table captions are too long and that text should be moved to the corresponding paragraph for further elaboration.
Answer: Done.
5.*****************
Continuing in Section 5, it is said that in a first step the way parameters vary with time is establish according to authors’ previous experience and won’t be discussed in the paper. In that case, if authors are not going to include a justification here, they should, at least, provide the functions ai(w) and bi(w) and include the corresponding reference.
Answer: Based on the experience with the analogue time-varying structures we assumed the exponential nature of the time-varying functions. Details regarding those functions can be found in [Osypiuk, R., Piskorowski, J., Kubus, D., A method of improving the dynamic response of 3D force/torque sensors. Mechanical Systems and Signal Processing]
6.****************
This tendency appears again in Section 6, where previous research is cited but no specific reference comes with that statement.
Answer: The conclusion on the duration of the parameters variation horizon comes from numerous our simulation experiments and were never discussed in a dedicated paper. In the future work we aim to propose some discussions on analytical proofs of given horizon values and proposed way of changing the parameters in time. Nonetheless, this seems not a trivial task and the research is still ongoing on our side.
7.****************
As previously said, this repeated lack of justification of assertions included in the text hinders to narrow down which proposals are original contributions that give rise to significant advances in the area.
Answer: The previous work that was used in this paper was mainly used as an experience in the domain of the time-varying systems. One can find our discussions in the previous papers about our growth in the subject but we didn’t want to overload the refences number with less significant publications that are less contributing to the currently present novelty method of reducing the transient time of the force/torque sensor measurement.
8.*****************
Coming back to Section 4, I suggest modifying Fig 2 so that sensor position and fixation method could be seen. With respect to Fig. 3, how was the settling time measured? According to the definition provided in tables1 and 2? If so, please consider moving the definition to Section 4.
Answer: Fig. 2 has been modified and now shows the sensor fixation method. Fig. 3 are views of the oscilloscope screen, which shows the voltage changes directly on the sensor, triggered by the capacitor discharge (left figure).
9.******************
Has the gun-based method been used before? Have the measurements been taken several times to ensure the repeatability of the experiment and the control of the impulse generation?
Answer: We have never heard of the use of gun-based method to determine the impulse responses of a force/torque sensor. It was an original idea that worked very well. The gun was fixed about 1m from the sensor.Multiple repetition of the experiment showed high repeatability of measurements.
10.*****************
Moving forward to Section 5, I suggest providing data about the HIL testing platform used. For example, about real-time processing and I/O options.
Answer: Done.
11.****************
In Section 6 authors establish that “In most test cases it allowed additional 10% improvement of the settling time in comparison to first stage optimization alone.” Please, elaborate more about the comparative improvement in all the cases and provide measurements and values. Besides, authors claim the superior performance for an impulse stimulus, what about the step response?
Answer: Table has been added that illustrates an exemplary results of the iterative optimization routine. One should remember that the mentioned iterative nature of the algorithm with each run the values will be slightly different.
As for the presentation only of the impulse response, we decided to highlight this characteristic due to more practical representation of the robot arm hitting an obstacle (hitting an operator).
12.****************
Regarding Fig. 9 and 10, further explanation is needed. Graphs are so small that it is difficult to extract conclusions. For example, does wo remain constant in Fig.9? and k in Fig 10? Either way, graphs should never be self-explanatory. Please, elaborate a bit more about parameter variations and explain titles (which is/are the current coefficients parameter?)
Answer: Figure 9 presents the first stage of the optimization routine (exemplary middle-stage of the iterative optimization algorithm). The first stage allows only linear change of the parameters in the given horizon (k, wo, zeta).
Figure 10 on the other hand allows nonlinear behavior of the parameter change according to the Bezier like curves. If the algorithm cannot find a better solution than in the first stage (linear behavior), some parameters might be still linear after that stage.
Other minor aspects:
13.**************
Please, put references in order of appearance: in line 90 references are not consecutive and, thus, references 20, 21, 23 and 24 appear in the text later than 22 and 25.
Answer: done.
14.**************
English needs revision, there are some articles missing and prepositions mistaken, as well as a couple of inconsistences between the plural and the singular forms of nouns/verbs.
Answer: We are trying our best. Some found mistakes were corrected.
Please, for better understanding, please rephrase/correct some sentences:
15.**************
Line 26: “Current measurement in the cascade robot control structure is one of key method used as indirect determination of forces and torques acting at a kinematic chain”
Answer: Done.
16.**************
Line 29: “Precise current measurements paved the way to force interactions with the environment,…”
Answer: Done.
17.**************
Line 36: “One of the approaches to a solution to this problem includes introduction of artificial compliance on the site of the robot or on the site of the environment.”
Answer: Done.
18.**************
Line 38: “Still, the most attractive approach includes lack of any required changes of the robot structure.”
Answer: Done.
19.**************
Line 51: “Based on JR3 sensor experimental environment has been prepared for verification purposes.”
Answer: Done.
20.**************
Line 55: “Properties of commercial filters implemented” What has been implemented? The properties or the commercial filters?
Answer: Done.
21.**************
Line 66: “A reduction in duration of transients in the filters applied would enable one to decrease the strength of the impulse-like forces.”
Answer: Done.
22.**************
Line 98: “Parameter h0 is calculated such that H(jw)max = 1.”
Answer: Done.
23.**************
Line 118: “Of course one can not use the transfer function representation for the time-varying systems.”
Answer: Done.
24.**************
Line 125: “The JR3 torque/force sensors clearly contributed to the development of commercial robotic systems interact with the environment.”
Answer: Done.
25.**************
Line 152: “Despite all measures, used, a special experiment was prepared in order to determine the impulse duration.”
Answer: Done.
26.**************
Line 169: “In this paper we investigated the possibility of designing and performance of the 500Hz and 125Hz filters (first and second in the cascade of hardware filters) as a digital time-varying structures.”
Answer: Done.
27.**************
Lines not provided (page 9): “We introduced the Bézier like curves as the base of the base of such non linearity.”
Answer: Done.
28.**************
Line 209: “The last point is given by the final, predetermined by the time-invariant design value. In this paper we run the random placement of the one middle control point.”
Answer: Done.
Sincerely yours,
Authors
Round 2
Reviewer 3 Report
Thank you for your answers and the corresponding modifications on the document. I consider that most of the issues I asked for clarification have been adequately justified.
However, there are a couple of minor aspects that I suggest further consideration.
Firstly, you establish that your main contribution is the proposal of a discrete time varying filter with parameter optimization, and that the analysis of the duration of the parameters variation horizon will be the core of another paper, still under development. I understand then that you won’t claim here its novelty, but in a future dedicated paper, as you say. However, in one of your answers you explain that the use of a gun-based method to determine impulse responses of a force/torque method is original. As I understand it, this is also an original contribution that deserves to be claimed as such in the document. For that reason, I suggest including, at least, a sentence transferring to the text the original character of the experimental method and its high repeatability, even with data, if available.
Secondly, and regarding Figures 9 and 10, I think that resizing the graphs has contributed to a better understanding of the linear and non-linear variation of the parameters, but I think that it would be better to also adjust the vertical axes to the real parameter variation ranges.
Finally, some possible typos:
Line 20: attention instead of attentions
Line 47: filter instead of filters
Line 48: its instead of their
Line 51: of the filter instead of of filter
Below equation 5, of the given structure instead of of given structure
Remove final dot at equation (7)
Line 167: digital 6th order filters or a digital 6th order filter instead of a digital 6’th order filters
Line 176: than instead of then
Line 184: in favor of instead of in favor to
The sentence “In this table: RiseTime - time it takes for the response to rise from 10% to 90% of the steady-state response, SettlingTime - time it takes for the error to fall to within 2% of steady-state response, SettlingMin - minimum value once the response has risen, SettlingMax -maximum value once the response has risen, Overshoot - percentage overshoot, PeakTime - time at which the peak value occurs.” Sounds weird and too “schematic”. Please, rephrase according to the style of the text.
Line 185: the definition of the Settling time has been already provided, so there is no need to include it again.
Line 219: I think that the expression “exemplary results” is not suitable here. Please, consider rephasing. For example: “As an example, Table 4 shows a parameter comparison using the optimization routine” or something similar.
Author Response
Dear Reviewer,
thank you very much for your comments. Your suggestions have been incorporated in the revised manuscript.
However, there are a couple of minor aspects that I suggest further consideration.
Firstly, you establish that your main contribution is the proposal of a discrete time varying filter with parameter optimization, and that the analysis of the duration of the parameters variation horizon will be the core of another paper, still under development. I understand then that you won’t claim here its novelty, but in a future dedicated paper, as you say. However, in one of your answers you explain that the use of a gun-based method to determine impulse responses of a force/torque method is original. As I understand it, this is also an original contribution that deserves to be claimed as such in the document. For that reason, I suggest including, at least, a sentence transferring to the text the original character of the experimental method and its high repeatability, even with data, if available.
Answer: I don't know if we've been precise enough in our last words. We are the authors of the idea of using CO2 gun to generate an impulse response of force/torque sensor. However, for the first time we showed this idea in an article: [Osypiuk, R., Piskorowski, J., Kubus, D., A method of improving the dynamic response of 3D force/torque sensors. Mechanical Systems and Signal Processing].
Secondly, and regarding Figures 9 and 10, I think that resizing the graphs has contributed to a better understanding of the linear and non-linear variation of the parameters, but I think that it would be better to also adjust the vertical axes to the real parameter variation ranges.
Answer: In the current example we decided to leave the vertical axes with current scale. If we plot the real possible ranges it might again disturb the reception and raise questions if the change of the particular parameter is linear or not.
Finally, some possible typos:
Line 20: attention instead of attentions
Answer: done.
Line 47: filter instead of filters
Answer: done.
Line 48: its instead of their
Answer: done.
Line 51: of the filter instead of of filter
Answer: done.
Below equation 5, of the given structure instead of of given structure
Remove final dot at equation (7)
Answer: done.
Line 167: digital 6th order filters or a digital 6th order filter instead of a digital 6’th order filters
Answer: done.
Line 176: than instead of then
Answer: done.
Line 184: in favor of instead of in favor to
Answer: done.
The sentence “In this table: RiseTime - time it takes for the response to rise from 10% to 90% of the steady-state response, SettlingTime - time it takes for the error to fall to within 2% of steady-state response, SettlingMin - minimum value once the response has risen, SettlingMax -maximum value once the response has risen, Overshoot - percentage overshoot, PeakTime - time at which the peak value occurs.” Sounds weird and too “schematic”. Please, rephrase according to the style of the text.
Answer: done.
Line 185: the definition of the Settling time has been already provided, so there is no need to include it again.
Answer: done.
Line 219: I think that the expression “exemplary results” is not suitable here. Please, consider rephasing. For example: “As an example, Table 4 shows a parameter comparison using the optimization routine” or something similar.
Answer: done.