SBD Dauntless, from scratch

[Wurger]

 

[Gnomey]

Nicely done so far!

 

[Witold Jaworski]

Wurger, Gnomey, thank you!

In this post I describe a break in the modeling that I made this week, because I had to fix my reference photos before the further work. The reason for this fixing was simple: the NACA cowling of my model did fit only the long-lens photos. For the further work I needed more information. This information was available in the high-resolution photos made by the Pacific Aviation Museum Pearl Harbor. However, they are slightly distorted.

In the 'mathematically ideal perspective' calculated for the computer cameras all of the straight lines remains straight. Unfortunately, the real-world camera lens can slightly deform (bend) the straight contours. This is so-called 'barrel' (or 'cushion') distortion of a photo. Unless you are using a panoramic lens, this deformation is hardly noticeable for the naked eye. Unfortunately, these differences become evident when you place a photo behind a 3D model, projected by a computer camera.

In case of reference photos that I used to verify my SBD Dauntless, the differences caused by the barrel distortion are visible around the forward part of the engine cowling:

It is really difficult to find the precise shape of this airplane on such a deformed photo. Thus I started searching the Internet for a method that would allow me to revert this deformation. First I encountered some advanced tools, like Hugin software. However, it seems to require series of similar photos to make a really improved picture. None of my single photos met this criteria.

On the Internet I also found some general scientific/engineering papers about the barrel distortion, as well as the tutorials how to fix it (for the architectural visualizations). They advised to use some originally straight contours/lines visible on the photo to estimate the image distortion. Indeed, for the photos as above, I can use the splits between airstrip slabs as such a reference:

I marked these lines on the photo above using dashed line. To see better their deformation, I draw along them straight lines, in blue. Note that these straight lines are tangent to the dashed lines on the left side of this picture. On the right side of this picture you can see maximum deformation of these split lines. This is the evidence of the barrel distortion.

I tried to reverse barrel deformation of this image using these split lines as indicators. I used a simple Lens Distortion filter from GIMP 2-D graphic program. The idea was that when I apply a deformation that makes these lines straight. Maybe such an operation will reverse the whole barrel distortion in this photo?

Figure below shows the Lens Distortion filter dialog window (GIMP), which I used to find the proper "reverse deformation" for this photo:

As you can see I used only the first (Main) parameter of this filter. I decreased its value until the split between the airstrip slabs on the preview became straight.

Then I matched the projection of my 3D model to this photo:

This is really a rough, approximate method, but the obtained results look really promising! Now the whole cowling fits the modified photo!

If it worked well for this picture, I tried it on another one:

The SBD fuselage spans over the whole length of this photo, thus the barrel distortion is more visible here. You can see it in the fin and the last bulkhead (see figure "a", above), as well as in the engine cowling (see figure "b", above). However, the contour of the hangar roof was a great reference for the reverse deformation. What's more, the GIMP dialog windows preserves the last used parameters. Thus I even did not have to adjust again the Lens Deformation filter! The same value of Main = -6.8 (as set in the filter dialog window - see the third figure in this post) made this hangar roof ideally straight. Both photos come from the same source, and their EXIF data reveal that they were made by the same camera. Thus I think that this deformation value is the "constant" property of this particular camera, which is repeated in each photo it made.

As you can see in figure above, after reverting the deformation, this picture perfectly matches the 3D model over the whole length, from the cowling to the fin.

During careful examination of all the nook and crannies of the fuselage, I encountered the difference at the root of the tailplane. In my model this rib seemed little bit shorter than in the photo (figure "a", below):

When all other element of this section fits the photo, such a gap means a real difference between my model and the original airplane. After some tweaks I decided that this rib was less deflected from the fuselage centerline (figure "b", above). (It seems that I made wrong estimation of its angle when I sketched these drawings). In fact, this rib run in parallel to the fuselage surface. Figure "c", above, shows that such a modified rib fits the reference photo pretty well.

As in the science: the theory is widely accepted, when it allows you to discover something previously unknown. These updated photos allowed me to find another error in my model!

I quickly converted most of the other photos from the same air museum. Unfortunately, I encountered the limits of this simplified method, when I tried it on the photos taken from a ¾ view:

I could not fit the model to the modified version of this photo! I had to revert to the original picture and its matching (using a slightly different lens length co compensate most of the barrel deformation).

It seems that the simple method of applying Lens Distortion deformation works only for the objects set in parallel to the picture plane.

For the consolation, I scanned again the Google image search (I have not done it for over four months). It was a fruitful idea, because I found two new reference photos, made by a long-lens (600 mm) camera. (They were published in this post from General Aviation News blog). The first of these pictures is even more banked aircraft than I have found before:

In spite of the same Navy blue camouflage, this is a different SBD-5 from Commemorative Air Force (note its '5' side number). It allows me to verify the important details of the vertical view: the width of the fuselage or the shape of the engine cowling. As you can see, the model fits this photo pretty well. In particular, I found here the confirmation of the new deflection angle of the tailplane root rib.

Another photo is an extremely high resolution (5400 x 3600 px) picture of the same aircraft, taken during landing. It allows me to check better the side view details:

Ultimately, on such a detailed picture I was able to find the dynamic deformation of the wing: it is slightly bent upward, so its tip is no more than an inch above its non-loaded location. The Dauntless wings were as stiff as in the fighters!

In this source *.blend file you can find one of these updated photos.

In the next post I will start to work on the NACA cowling details, using the reference objects formed in the two previous posts.

 

[Wurger]

 

[Gnomey]

Good work so far!

 

[Witold Jaworski]

Wurger, Gnomey, thank you!

In this post I will shape panels of the Dauntless NACA cowling. Working on the scale plans a couple months ago I came to the conclusion that the basic shape of this cowling was the same in all the SBD versions (see Figure 4.6 in this post). You can find the differences in their 'ornaments', like the sizes and locations of the carburetor air intake, or the number of their cowling flaps. Thus I used the high-resolution, long-lens photo of the SBD-5 (described in the previous week), to determine the ultimate shape of this cowling, and the split lines of its panels (figure "a", below):

Basically, the SBD Dauntless NACA cowling was split into a single upper panel and two symmetric side panels. I started by copying corresponding part of the reference shape (created in this post) into the single side panel (figure "b", above). The subdivision surface of such a 120⁰ mesh 'arc' is somewhat flat at both ends. Thus I had to tweak a little mesh edges in these areas, fitting them to the reference contour.

In the next step I extruded the 'strip' that overlapped the upper panel (figure "a", below):

I also marked the bottom edge of this panel as sharp (figure "b", above). In fact, the right panel overlapped the left panel along this line (they were similar, but not identical).

What's more, in the SBD-5 and -6 the split line between these panels was shifted left by about one inch. Nevertheless I decided that I will split these two panels later, during the detailing phase. At this moment I just dynamically mirrored the left panel using modifiers. It will be easier to unwrap in the UV space this single element, then copy its unwrapped mesh and form the right panel during the detailing phase.

To keep the topology of this mesh as simple as possible, I decided to cut out the exhaust stacks openings using a Boolean modifier:

The high-resolution photo was a very useful reference for the ultimate check of the shape of this opening. (Its contour contains two arches connected by short straight lines).

In a similar way I cut out the space for the cowling flaps:

Actually, I am preparing the three-flaps sections, as used in the SBD-1.. -4. Note that I used the same auxiliary object to cut the upper cowling panel.

The overlapping 'strip' along the upper edges of the side panels was chamfered just on the cowling leading edge (figure "a", below):

It would be very difficult to shape such an effect 'in the mesh' here, because of the two-dimensional curvature of this area. That's why I created it using two auxiliary objects and another Boolean modifier (as in figure "b", above). This was the last detail of this panel, for the modeling phase.

The next element are the cowling flaps. Initially I created them as a three-segment 'strip' (one quad face per each flap). I marked all edges of this initial mesh as 'sharp' (Crease = 1). Once I determined the size and shape of these basic faces, I added new, internal edges and started to bend this 'strip' along the reference shape (red object in the figure below):

When this 'strip' was fitted to the reference cowling panel, I added temporary edges connecting their opposite vertices. These auxiliary lines helped me to determine direction of individual rotation axes of these flaps, as well as their origins (figure "a", below):

Then I separated appropriate fragments of this mesh into three cowling flaps (figure "b", above).

Finally I cloned and mirrored the three left cowling flaps into the three right cowling flaps (figure "a", below):

At this moment the right flaps objects have a negative scale, thus for the movement test I have to rotate these left and right flaps separately (along their local Z axes, using the Individual Centers pivot point mode — as in figure "b", above).

In this source *.blend file you can evaluate yourself the model from this post.

In the next post I will form the gun recesses in the upper cowling panel. It will be a quite difficult detail!

 

[Wurger]

Great !

 

[Gnomey]

Lovely work so far!

 

[Witold Jaworski]

Wurger, Gnomey, thank you!

The gun recesses in the aircraft usually are tricky elements. Their edges depends on the shape of two curved surfaces: the fuselage around the recess and the tubular inner surface. When you make mistake in any of these two shapes — you have to remodel the whole thing.

In the SBD there are two symmetric gun recesses in the upper part of the NACA cowling. Figure below shows the left one:

As you can see on the photo, these recesses were formed in a separate metal sheet. It was riveted afterwards to the main body of the NACA cowling. I will repeat such an arrangement in my model, because using a separate object for such a feature simplifies its mesh. (I can make this mesh denser than the NACA cowling around it, and still I do not have to worry about the topological implications). The sheet metal around these recesses seems to be relatively thick, which ultimately makes the fitting of this panel to the NACA cowling surface easier. To make some space for this dedicated panel, I created initial openings for the gun recesses in the upper panel of the NACA cowling. They are generated by a Boolean modifier, and are a little bit larger than the final recesses.

The most difficult part of the gun recess in this aircraft is the fillet around its edge. To obtain a high-quality shape, I decided to start this panel as two separate objects. The first of them is the tubular inner surface (copied from the "cutting" object used in the Boolean modifier). The second object is just a small cylinder, which radius is close to the fillet radius. I will deform it along a 3D curve, which follows the border of the gun recess opening:

When I started to extrude subsequent segments of the "fillet" cylinder, it automatically follows the assigned curve. (The curve allows me to do it without worrying about preserving the circular cross-section along the whole length of the opening border). Technically, this is the effect of a Curve Deform modifier that I assigned to the cylinder object. This is the first modifier in the stack, and it precedes the smoothing (Subdivision Surface) modifier. Such an arrangement allows me freely slide the circular cylinder sections around the opening border, finding the proper locations for these key vertices:

Then I shifted this resulting contour down, and adjusted its "spine" curve so that the "fillet" cylinder barely touches the opening edge.

When the basic cylinder was shaped, I removed (applied) the curve modifier, as well as the unnecessary ¾ of the cylinder surface. The result is a regular fillet, formed around the opening (figure "a", below):

Now I started prepare the inner part of this recess for joining with this fillet. I had to add some additional sections. They are placed at the corresponding sections in the fillet mesh (figure "b", above).

When all the edges of the inner recess mesh were verified and adjusted to match the filet, I joined these two objects and removed the unnecessary faces (figure "a", below):

Then I created new faces that join these two meshes (figure "b", above).

Once the inner part of the recess panel was completed, I started to form its outer part by extruding its outer edge (figure "a", below):

I placed its vertices on the outer edges of this panel (figure "b", above). Then I added another edge loop in the middle and started to elevate the 'sunken' part of this surface above the cowling panel (figure "c", above).

Figure "a" below shows the outer surface neatly fitted to the cowling. As you can see, it requires not one, but two inner edge loops outside the fillet, to reproduce circular cross section of the NACA cowling around this recess:

Finally I used the same auxiliary object as for the underlying panel to cut out the space for the topmost cowling flap (figure "b", above). (It is made using the Boolean modifier).

The gun recess in figure "b", above, looks good enough. However, when I looked onto another reference photo, and then onto another, I slowly started to discover that these recesses had different cross sections! I assumed that it was an arc, while the more I study the photos, the more I came to a conclusion that it had narrower, 'U'-shape cross-section!

Such surprises are common, when you are making a precise model. Thus, do not assume that the progress of your work will go as a "waterfall". It is more similar to a "spiral": you often come back to the completed parts and adjust some of their details. Just keep the objects ready for such situations: they are normal part of the work.

That's why I still keep as much features as possible implemented as the modifiers applied to relatively simple meshes. Thanks to such an arrangement, the adjustment of the recess shape does not require a lot of work:

First I created a simple auxiliary object as the reference of the correct cross-section shape (white contour in figure "a", above). Then I placed the panel being modified over the reference shape of the NACA cowling (in red). Then I started to shift the complete fillet sections and the near lengthwise edges in the front view, placing them on the new contour. When it was done, I made minor adjustments along the recess edge, shifting the fillet sections until they fit the red surface of the NACA cowling.

The difference in colors helps me to estimate the remaining deviations from the reference surface. I usually shift the modified section downward, until the resulting gray surface around it 'sinks' in the red reference surface. Then I move it minimally upward, so that the resulting surface appears just above the reference object.

Figure below shows the final result: gun recesses in the upper panel of the NACA cowling:

For convenient "handling", the gun recess panels are attached to their cowling panel by the "parent" relation in the internal hierarchy of this model.

In this source *.blend file you can evaluate yourself the model from this post.

In the next post I will form another element of the upper cowling panel: the carburetor scoop.

 

[Wurger]

 

[Gnomey]

Good work so far!

 

[Witold Jaworski]

Wurger, Gnomey, thank you!

This week I have worked on the carburetor air scoop. This scoop passed significant evolution in the subsequent Dauntless versions. In the SBD-1 there was a rather large air duct placed on the top of the NACA cowling (see figure "a", below):

However, it was quickly discovered that it obscures one of the most important spots in the pilot's field of view: straight ahead and slightly below the flight path. That's why it was somewhat corrected in the next version (SBD-2). In this aircraft the designers lowered the scoop, increasing the field of view from the cockpit (see figure "b", above). Such a solution persisted in the SBD-3 and -4. In the SBD-5 they completely redesigned it, placing the carburetor scoops inside the NACA cowling (more about this — see in this post the paragraphs around Figure 11-6).

Close examination of the various reference photos led me to the conclusion that in the SBD-1 the air duct ran between the inner surfaces of the scoop and the top of the NACA cowling (figure "a", below):

There was a rectangular opening in the rear part of the cowling, located just above the Bendix-Stromberg carburetor of the R-1820 engine. (There was a short, vertical duct inside the NACA cowling from this opening to the carburetor intake. I will model it later, together with the engine).

The later scoop version (from the SBD-2, through SBD-3, up to SBD-4) was a typical "quick and dirty" solution for the identified problem. The designers could not split the upper panel to place the lowered air duct there, because it would hinder the stiffness of the whole NACA cowling. Instead, they cut out another rectangular opening in its leading edge (figure "b", above). In this way a half of the incoming air went to the engine as before, over the NACA cowling. However, the bottom part of the air stream was directed below the cowling surface. Both streams were joining inside the rear opening, before they went into the carburetor.

I created both openings using Boolean modifiers:

Then I started by forming the lower part of the air intake. I started with a single strip fitted to the side edges of the frontal opening (figure "a", below):

Then I extruded this edge and flatten the subsequent segments, forming the characteristic shape of the inner inlet, as in the reference photos (figure 'b", above).

When this first part of the bottom air duct was ready, I extruded its subsequent segments, forming the rear part (figure "a", below):

Finally I reduced the roundings along the duct side edges by adding there a multi-segment Bevel modifier. It not only diminished their size, but also made its cross section more circular (figure "b", above).

When the bottom part of the scoop was ready, I started the upper part. It begins in the same way: from a single strip, fitted to the cowling surface (figure "a", below):

Then I extruded the vertical faces (figure "b", above).

In the next step I extruded their upper edge into the horizontal surface (figure "a", below):

Finally I extruded the subsequent segments of the rear part of this mesh (figure "b", above).

Initially I kept the lengthwise edges of this object sharp, because I intended to create their fillets using the Bevel modifiers. However, a careful study of the reference photos revealed that the radii of the upper and bottom edge vary along the length of the scoop. Thus I created them by adding two additional lengthwise edgeloops to this mesh:

Figure below shows the real scoop (on the left) and the final version of the same scoop my model (on the right):

Although I did not managed to set up the picture on the right precisely as in the left photo, the carburetor scoop looks quite similar on both images. I can leave it "as it is" and start the work on the next cowling element. I can always fix its shape during the next stages of this project.

In this source *.blend file you can evaluate yourself the model from this post.

 

[Wurger]

Great!

 

[VBF-13]

Witold, I've never seen anything like this in my life. My Dad carrier-qualified in a "Speedy-D." This is truly fascinating. Thank you so much for this effort.

 

[Lucky13]

Absolutely fantastic work!!

 

[Gnomey]

Lovely work so far!

 

[SANCER]

It's been some time since the last time I was here.
My dear Witold, your job is simply AMAZING !!!
The level of detail and perfection that you used is Intimidating !!

Saludos y felicidades amigo
Luis Carlos

 

[Witold Jaworski]

Wurger, VBF-13, Lucky13, Gnomey, SANCER - thank you!
VBF-13 : It's my pleasure! I like this aircraft since I was boy. From the veterans' memoirs it seems that it was like a good soldier - you could trust it in action.
SANCER - many thanks for my first rating

In this post I will finish the engine cowling of the Dauntless (of course, for this stage of the project). In the previous posts I formed its outer panels. In the case of the air-cooled radial engines like the one used in the SBD, there is always another, inner panel: the central part of the cowling. It is located behind the cylinders and exhaust stacks. In the classic arrangement of the NACA cowling it is nearly invisible. In the SBD-1..-4 you could see only its outer rim. That's why I had to use all available pictures of the Dauntless engine maintenance or the wrecks, to learn about its general shape:

This panel had two variants. The first one (let's call it "flat") is visible on the photo above. It was used in the SBD-1..-4. In the SBD-5 and -6 the engine was shifted forward by 4", so the central panel became a little bit longer ("deeper").

Frankly speaking, I still need more photos and drawings to better determine the shape of this part, especially the details of its earlier, "flat" version! Let me know if you have one — I am especially interested in the upper area, around the carburetor, in the SBD-1…-4. (The few photos that I have reveal that behind the upper cylinders of the R-1820 engine there was a vertical air duct from the air scoop to the carburetor. I still need to determine its shape, as well as the shape of the inner cowling around it).

That's why I decided to determine the exact shape of this hidden panel later, when I fit the engine and its mounts. (I count on the indirect information coming from the geometry of the engine mount and the exhaust stack shape). At this moment I am leaving this area "as it is", because too much of its geometry is based on my assumptions.

However, I can precisely shape the recesses around the gun barrels, because they are better visible on the photos. I have to make these details easily adaptable when I have to alter the shape of this panel. (I expect that in the future I will tweak the area around the carburetor multiple times, before it "stabilizes" in the most probable state).

The cross-section of these gun recesses have the same shape as their troughs in the NACA cowling. Thus I started by copying the control polygon of this "U"-like cross-section shape (five control vertices) and extruding it into an auxiliary "trough" (see figure "a", below):

I examined the interesection edge of this auxiliary object with the central panel. The goal was to place its vertices as close as possible to the existing mesh edges. I could easily check it in the front view, because the "trough" in this projection is reduced to a single contour (figure "b", above). While the both of its side vertices are very close to one of the elliptical edge loops, the middle vertex was too far from the nearest radial edge loop. I had to adjust the mesh of the central panel by rotating a little all of its upper radial edges.

After these preparations, I generated in the panel mesh the intersection edge with the auxiliary "trough" (I used my Interesct add-on for this purpose):

I removed the three vertices that were inside the contour of this intersection. It also deleted all the mesh faces around these points. Then I created new faces in this place, merging the intersection contour with the rest of the mesh of this panel (as in the figure above).

Figure below shows how I created the inner surface of this gun trough:

I started by creating a new face that "bridged" the opposite edges of the opening. Then I split it twice, obtaining three inner edges. I placed these edges directly behind the corresponding vertices of the opening contour. Then I closed this opening, creating the four remaining faces. (Now I can see that I could do the same in a simpler way, by extruding the bottom part of the opening contour. Never mind, both methods lead to the same result). At this moment the edges of this recess are too smooth. To reduce the radius of this rounding, and make it similar to a regular fillet, I assigned these edges the full Bevel Weight (=1.0). Then I added to this object a multi-segment Bevel modifier (before the smoothing Subdivision Surface modifier). The last picture from figure above shows the faces generated by this Bevel, before they were smoothed.

Finally I compared the shape of the resulting gun trough to the corresponding troughs in the upper cowling panel:

(I made it transparent, to better see the eventual differences in their shapes). Indeed, there were some deviations. I quickly fixed them, adjusting in the front view the whole edges of this recess. (In this view these edges are reduced to a single point).

Now I have to trim ends of the troughs in the NACA cowling, creating the space for the central cowling panel. I could do it by modifying their mesh. However, because the shape of this panel may be altered in the future, I decided to use another Boolean modifier for this purpose. I just created an appropriate auxiliary object, and applied it to the gun trough panel:

This was the last element of the NACA cowling. Figure "a", below, shows the recesses in the central panel that I formed in this post, while figure "b" shows details of the whole assembly:

As you have probably noticed in the course of the few previous posts, I had often to move the location of the NACA cowling, switching between the SBD-5 and the SBD-3 versions. To avoid such endless movements in the future, I decided to split the Bledner file of this project into several separate scenes for each Dauntless version that I need. For the beginning I created two additional scenes, for the SBD-1 and SBD-5. They are named after the Dauntless version they contain, thus I renamed the current scene to "SBD-3".

Figure below shows the SBD-5 scene (and the scene selection menu):

When I created the scene for this Dauntless variant, I chose the option that created it as a copy of the original scene. Initially both scenes share the same objects (the same fuselage object or the wing objects are "linked" to SBD-3 and SBD-5 scenes). In the effect, I can edit these shared objects in any of these scenes. Every change I apply to their meshes, modifier stacks, or general positions/scales/rotations is visible everywhere.

Because the NACA cowling in the SBD-5 was shifted forward by 4", I had to make in its scene local copies of the panel objects. However, they still share with the SBD-3 their meshes. In the effect, they became "clones" of their counterparts from the SBD-3 scene. Clones share the common meshes, thus they have the same basic shape, but they can have different general transformation (location/rotation/scale). Thanks to this, in the SBD-5 scene the bottom panels of the NACA cowling have the same shape as in the SBD-3, but their location is different. What's more, the clones can have different modifier stacks. Thus in this SBD-5 model I was able to remove the carburetor scoop openings from the upper NACA panel, and modify the cutouts for the different cowling flaps (see figure above) because they were generated dynamically, by a Boolean modifier.

Ultimately — there are a few objects specific for the SBD-5, which exist only in this scene: the central cowling panel and the panels around the gun troughs. I copied their meshes from the SBD-3 and then modified them according the SBD-5 reference drawings. In the SBD-5 the central cowling panel, placed behind the engine cylinders, was longer by 3.5" than in the previous versions. I had to scale and reshape this mesh. Fortunately, its gun recesses (formed at beginning of this post) are easily adjustable, thanks to their simple topology.

In similar way I created a separate scene for the SBD-1:

At this moment the only difference between the SBD-1 and SBD-3 is the carburetor scoop on the top of the NACA cowling. However, there will be another minor differences in the next row of cowling panels.

In the future I will also create the SBD-2 scene (combining the NACA cowling from the SBD-3 and further cowling panels from the SBD-1), and the SBD-4 scene (basically – it is the SBD-3 with the SBD-5 Hamilton Standard Hydromatic propeller). As you can see, the SBD-2 and SBD-4 will be just combinations of various parts from the "key" versions (SBD-1, SBD-3, SBD-5), thus I will create them at the end of this build.

In this source *.blend file you can evaluate yourself these SBD-1, SBD-3 and SBD-5 scenes and their initial contents. In the next posts I will continue my work on the SBD-3, then update the SBD-1 and SBD-5.

 

[Wurger]

 

[Gnomey]

Looking good so far!