Data-Driven Modeling of Fuel Consumption for Turboprop-Powered Civil Airliners

Round 1

Reviewer 1 Report

This paper proposed a data-driven response surface model for fuel consumption in Civil Airliners. A systematic data set are provided for training the surrogate model and validating the performance. The data-driven model shows an accurate prediction and therefore can be used for the initial design of turbo-propeller aircraft.  Based on this data-driven modeling, the design cycle can be significantly accelerated. The authors illustrated this method using several examples with clear datasets. 

This paper is well structured and organized. The idea is novel for specific engineering applications. I recommend it is accepted for publication.  

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Data-driven Modeling of Fuel Consumption for Turboprop-powered Civil Airliners

In this manuscript, a response surface model for estimation of fuel consumption as a function of payload and range of a flight is presented. This model is proposed as a variant to Breguet equation.

In general, problem statement in the introduction could have been stronger by explaining the difficulties in estimation of accurate fuel consumption by using alternative approaches.  It should have been also discussed the cause of the error and its magnitude in those models. Then, the proposed model and the niche it will address could have been added clearly. Moreover, there are some assumptions such as constant cruise speed and constant engine temperature (ultimately propulsion efficiency) that affect the validity of the proposed model unless it is shown that those parameters do not vary in real conditions as well. Otherwise, the title, abstract and conclusion should explicitly explain the scope of the model and its uncertainties, due to those simplifications. The last general concern is the
data-driven modeling” which is in the title of the article might be misleading nowadays since only 16 samples seems to be considered (table 1 and the manuscript suggest 16 samples only).

In conclusion, reviewer suggests a major review for the article before considered for publication.

The specific comments are as follows:

• Page 2 Lines 35-45

Second paragraph of the introduction, where literature review is supposed to come based on the manuscript’s structure, is vague and doesn’t give a clue of what are the characteristics of the models in the literature, their advantage and disadvantages and ultimately the reason and the niche for conducting the current research.

• Page 2 Lines 35-36

18 references are brought without any explanation on the importance of each. Explanation is required for every reference and its relevance to the current topic.

• Page 2 Lines 40

“It is often, …”

What is often assumed to be constant?

• Page 2 Lines 41-43

“But the estimates……to large discrepancies.”

 Aerodynamic parameters can be calculated by experimental and numerical tests. While they are variable as a function of airplane speed, airfoil shape, etc., some factors affecting this variation such as airplane speed are neglected since they are considered constant (Page 2, line 62: “max. cruise speed is used..”).

Moreover, propulsion and its relation to mass flow rate of fuel depends on the drag exerted on the plane, temperature of engine that affects engine efficiency, etc., which are considered constant.

• Page 2 Lines 46-55

It is hard to follow paragraph three of the introduction.

• Page 2 Line 56

Adding the Breguet equation to the beginning of section 2, along with the clarififcation on its terms and the ones with large uncertainty that have been addressed in the current study would be significantly beneficial to audience of the journal.

It would also be beneficial to add a flowchart that summarizes the steps taken in this somewhere under section 2.

• Page 2 Lines 58

It is recommended to start the explanation of the dataset with what it is, who gathered it (reference), how many samples are included, what types of airplanes are considered, and then also elaborating on how splitting was was performed?

• Page 2 Lines 59

“…comprising nine types from eight different manufacturers.”

            Nine types of what? Please clarify.

• Page 2 Line 62

Considering the maximum cruise speed and ignoring one crucial parameter in fuel consumption should be justified by referencing to studies that show that significant duration of a flight is at cruise speed with low variations. Otherwise this important assumption should be reflected in the abstract or probably the title.

• Page 2 Line 68

What is BADA database? Since energies is not only a flight journal, explanation is required for the audience not expert in this field.

• Page 3 Table 1

This table shows that data is made up of only 16 samples so calling it a data-driven analysis may not be appropriate nowadays when it data-driven is referred to large set of data in order of 1000 and more.

• Page 3 and 4

Equations 1-4 are gathered from the literature and should be referred to the literature in the manuscript.

• Page 4 Line 93

What is SR? it is not defined.

• Page 5 Line 130

It is recommended to add a description of the effect of each p-coefficient, since it is focused on later on in Figs. 3-4.

• Page 5 Line 139

“…in Fig. 4…”

• Page 5 Fig. 3

Coefficient p₂₀ should be multiplied by square of the range (Ra²) as shown in equation 6.

• Page 6 Line 181

The reference of previous trend is [22] in the manuscript but in Fig. 5, the previous fit is referred to as reference [1]. If they are different, please clarify it in the manuscript. If they are not, please correct the reference number.

• Page 8 Fig. 7

Horizontal axis of Fig. 6 is Gross Weight GW while in Figs. 5 and 8, GW is used alone. Please adjust to be consistent.

• Page 8 Lines 200

“We also plot… “

Specific Fuel Consumption is the only parameter that is mentioned. Please add the Gross Weight to clarify what can be seen in Fig. 7.

• Page 8 Lines 205-206

“This confirms that the power…engine power [4].”

How the expression above confirms the observations

• Page 10 Lines 252-256

As is also observed in Breguet equation, propulsion and its efficiency affect fuel flow. Ignoring this effect affects the final results and their validity. This significant assumption should be mentioned in abstract, conclusion and title of the paper, to clarifies what the model misses.

Comments for author File: Comments.pdf

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

The work is an extension of the research presented in the previous publication of the authors [22].
The differences are in using a different and larger set of aircraft training and testing data.
Previously, an insufficient (5 aircraft) data set was used for statistical sense. It is assumed that the minimum data set is 6. Now 9 planes have been used. The validation was previously carried out using only one example.
Previously, data were related to high-mass military aircraft. Data for civil aircraft have now been used.
The methodology for building the fuel consumption model is identical to the one used previously.
The authors have now used a slightly larger (9) than the minimum (6) set of cases for statistical studies.
According to the data contained in Table 1, nine training and six test data sets were used.
The model was developed under certain assumptions.
Assumptions seem reasonable.
In the previous work the model was presented and validation was carried out using only one test example. In the current work, 6 aircraft were used to validate the new set of model coefficients.
The use of a new and key assumption is important in both works (old and current)
that p_ij coefficients can be described by a polynomial with an OEW parameter.

I would put forward the thesis, that the lack of matching factors to some test data may be a measure of the structural correctness of these aircraft.

It is extremely important that the model requires only one parameter - the weight of the aircraft.

One of the unexpected conclusions is the ability to calculate the amount of fuel needed to reach operating altitude based on the airplane weight (OEW) rather than take-off weight.

It seems appropriate to extend the test database beyond the ones used 15.

Specific comments.
Fig. 5 no scale on both axes. Request to add an explanation for the reason for the lack of scale. Planes are civil. In a previous article by authors, on a similar chart for military aircraft, the scale was placed.
The same applies to fig. 6 and partly fig. 7, as well as fig. 1.

The authors proved that their method gives 2-3 times smaller error in calculating fuel consumption in relation to the Breugeot equation.

The work is a development of the method discussed in the previous article of the authors but the thematic scope is different (relatively light civil aircraft now, heavy military aircraft earlier) and the current work is based on more data, training and testing, which means that the results are much better documented.
It would be suggested to respond to the correctness of the assumptions under which the calculations were carried out, when their results are known.

The work may be published after minor corrections suggested in the review.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Thanks to authors for their effort to improve the paper, and a great job adding fig 5. I still have 3 concerns about the article before publication:

• In paragraph 2 of the introduction it is mentioned that several models… which shows that there are several significant models that should be described and what are their disadvantage that led to current paper.
• SR above eq. 4, which comment 13 of the original review is not defined yet.
• Since temperature is constant, it should be mentioned explicitly in the abstract while the range of propulsive efficiency is used only.

Author Response

• In paragraph 2 of the introduction it is mentioned that several models… which shows that there are several significant models that should be described and what are their disadvantage that led to current paper.
  > Except for the Breguet equation, there is to the authors' knowledge no other alternative in the literature. Paragraph 2 rephrased to better highlight the existing alternatives:
  Several variants are available to make operating empty weight and fuel weight predictions that can be applied to diverse aircraft types: [2-19] to cite just a few. Either actual data is given about a few existing designs or the Breguet range equation [20] is invoked with the fuel fraction method to estimate the fuel burned in cruise conditions for similar designs.
  > the pros and cons of the Breguet method are discussed in that paragraph.
• SR above eq. 4, which comment 13 of the original review is not defined yet.
  > SR defined at line 70, page 2
  > as requested, line 106, page 4, rephrased:
  ... from interpolated specific range SR(GW,Alt)-relationships...
• Since temperature is constant, it should be mentioned explicitly in the abstract while the range of propulsive efficiency is used only.
  > Referring to comment 4 of the original review, the authors' cannot comply with this request since it would be false to pretend working under any 'constant temperature' or 'constant engine efficiency' assumption.
  Engine parameters vary along the course of a flight. So does the engine temperature and the engine efficiency. What the method does is to look beyond those natural fluctuations, and so works the Breguet method too.
  Let us look at an example to clarify:
  Say a driver drives from home (where he ignites the car by +15°C) and drives a long way North 1000km far where the temperature gradually decreases along the trip to reach -5°C locally. Keeping track of the fuel he/she filled up, the driver is able to come with an averaged statistic of XX L/100km. But this statistic is in no condition under any constant temperature assumption since the engine had to warm up at the start, went through many temperature cycles with the cooling firing up, and finishes at high efficiency given favorable external conditions. Not even to speak about the many stops and gos, climbs and descents, ... affecting the engine's instantaneous efficiency.
  The abstract is clear about the response surface method (which is a form of averaging), on the fact that we predict fuel weight for a mission (ie. payload and range pair) and thus not an instantaneous consumption, and ---thanks to the valuable reviewer's comment--- that the work is under the current range of efficiency of the propulsion ---should any breakthrough yield new horizons in terms of efficiency, the method cannot be applied and should be updated---.
  > The influence of temperature deviation from the international standard atmosphere are dealt with in section 3.3.

The uploaded manuscript has been edited with track changes on to highlight the current changes.

Author Response File: Author Response.docx

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.