Online Estimation of Three-Directional Tire Forces Based on a Self-Organizing Neural Network
Round 1
Reviewer 1 Report (Previous Reviewer 3)
the paper can be now accepted for publication, in the reviewer's opinion
Author Response
Thank you very much for your excellent and professional revision of our manuscript.
Reviewer 2 Report (New Reviewer)
General comments:
1. An interesting topic with relevance to modern, advanced vehicle control systems
Editing:
1. Paper is generally readable, but I recommend a native English speaker with a technical background proofread this paper.
2. Some figures, especially in the results section, need to be improved as they are unclear. See comments in this PDF.
3. Some paragraphs are highlighted yellow, some with red text. It is unclear why this has been done.
Technical
1. An online tire friction and force estimator is presented. This is a relevant topic, especially for autonomous and semi-autonomous vehicles. It is a well-known and deeply studied problem.
2. The study makes use of a highly simplified tire model that does not accurately capture the relationship between slip angle, slip rate and tire force.
3. The study also refers to estimating the 'adhesion' coefficient. It is unclear if this is the peak friction coefficient or the current friction coefficient between the tires and the road surface. Estimation of the peak friction coefficient is valuable from a vehicle safety perspective but has limited use for estimating current tire forces.
4. The lateral stability is sometimes referred to, but this adds limited value to the research question. The only real contribution including the LTR is to indicate that the vehicle is mostly operated within the stable, linear region. Because of this, the results are not representative of the full range of vehicle operation and lack a degree of robustness.
5. The initial estimate of friction will play a significant role in the result. It is unclear how this will affect the proposed approach.
6. In the results section, the error is compared to 'real values'. Are these measured tire forces? How were these values obtained?
Comments for author File: 
Comments.pdf
Author Response
Dear Machines editors and Reviewer,
Thank you very much for allowing me to submit this revised manuscript. This paper has been edited by a native English speaker before submission and has an editor's certificate. A professor with technical background has proofread this paper again in order to express more professionally. We describe the recognition process of the road type classifier in more detail. To make the results displayed more clearly, in section 4.2, we updated the result plots and bolstered lines and enlarged some of them.
Appreciate to you for considering the possibility of publishing this article in machines.
Yours Sincerely,
Shaohua Li
Author Response File: Author Response.pdf
This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.
Round 1
Reviewer 1 Report
This paper discusses a Levenberg Marquarelt multi-13 module and self-organizing feedforward neural networks (LM-MMSOFNN) model for tire forces estimation for vehicle dynamics. The authors presented both simulation and experimental validations of the proposed LM-MMSOFNN methods. The paper is clearly written and this reviewer has the following comments and suggestions.
1. The motivation of the work should be further clarified. The approach takes the vehicle dynamics model and a tire friction force model for the LM-MMSOFNN model; see Fig. 8. It is unclear how and why the LM-MMSOFNN is developed if the vehicle dynamics (including tire dynamics) models are used. The vehicle dynamics model can be used to predict the frictional forces and road surface conditions. Does the new neural network model provide any additional information or more accurate? If so, the paper should give a clear description and validation results on this.
Second, the authors should discuss more about why the LM-MMSOFNN is used (not other models) for the proposed systems. Also, the authors should discuss clearly how to train the LM-MMSOFNN model. Right now, the discussion of training is very limited. The paper should also give a clear validation of the new learning model for road conditions that are not among the training data set. This is critical to demonstrate the performance of the new model.
Finally, it is unclear whether the proposed neural network models can be used for real-time applications. The paper does not clearly discuss this and these figures do not explicitly present whether the data are obtained in real-time. If so, the paper should discuss the computational or complexity analysis of the learning model approach.
2. Although the paper includes many important literature on tire friction models and embedded time sensors, the authors might still need to include some important developments in the high-relevant topics. For example, adaptive control or estimation with analytical tire model could be used for identifying the road conditions or friction coefficients, to name a reference such as Yi et al. (2002). Adaptive emergency brake control with underestimation of friction coefficient. IEEE Trans. on Control Systems Technology, 10(3): 381-392, etc.
Author Response
Thank you very much for your excellent and professional revision of our manuscript. We made a detailed reply in the response and revised it in the manuscript.
Author Response File: Author Response.pdf
Reviewer 2 Report
Before I can comment on this paper the authors should revide the vehicle model by taking into account the roll stiffness/load transfer distribution. Nomenclature is confusing and requires rework.
Most of all, an accurate list of input signals, including where they come from, is strongly required.
Author Response
Thank you very much for your excellent and professional revision of our manuscript. We made a detailed reply in the response and revised it in the manuscript.
Author Response File: Author Response.pdf
Reviewer 3 Report
the paper relates an interesting topic, since the evaluation of the interaction forces and in particular of adherence is a target crucial for vehicle dynamics and automotive market, in particular in the evolution of the autonomous and connected vehicles
despite this, the research shows deep lackings in its vehicle dynamics parts. it seems the authors are expert in data processing and machine learning but not much involved in tires and vehicles dynamics, fundamental in such kind of paper
in particular, the main issues regard the complete absence of information regarding the identification of the tire model from the
bench experimental data (that, moreover, report curves in figures 3 and 4 characterized by differences from load and unload phases, that cannot be reproduced by a model like dugoff or pacejka. for this reason, much more should be said on how the tire model has been parameterized and implemented
from the methodological point of view, the proposed case study is quite poor, because it is simply represented by the training of a Neural Network on a set of data generated by a commercial software superimposing several levels of "friction" (which is just of the interaction forces scaling factor), that are then recognized as expected by a machine learning algorithm, without any further validation outside of the training ranges
as regards the paper results, it is not clear how the experimental adherence value of 0.85 has been obtained, and many doubts are related to the absence of sensors able to directly measure the tire forces, as dyno hubs.
due to their absence, it is not clear how the estimated tire forces can be compared to "actual" longitudinal, lateral and vertical ones as done in figures 18, 19 and 20
in the reviewer's opinion, the authors should be much more deeply satisfy several conceptual aspects, discussed in a too generic way, before the paper can be published
Round 2
Reviewer 1 Report
The authors addressed my previous comments and concerns and I do not have any further concerns.
Author Response
Thanks very much for your contribution to the improvement of the paper.
Reviewer 2 Report
I do not see significant improvements over the first version. Roll stiffness/load transfer distribution is still neglected. Still it is not clear how the vehicle speed is retrieved or computed.
Author Response
We would love to thank you for allowing us to resubmit a revised copy of the manuscript and we highly appreciate your time and consideration.
Author Response File: Author Response.pdf
