Mathematical Foundations and Computer Graphics

Introduction

Computer-aided design (CAD) and computer-aided manufacturing (CAM) depend on a precise mathematical description of geometry. A CAD system must represent the position, orientation, size, shape and topology of engineering objects; a CAM system must then interpret that representation to generate tool paths, verify machining operations and communicate with manufacturing equipment. Mathematical foundations therefore provide the bridge between design intent and physical production.

This chapter develops the principal mathematical tools used in CAD/CAM: coordinate systems, vectors, matrices, transformations, projections, clipping and visualization. It then examines mathematical representations of lines, curves and surfaces, with particular attention to Bézier, B-spline and non-uniform rational B-spline (NURBS) geometry.

Learning Objectives

By the end of this chapter, the reader should be able to:

• distinguish world, model, view and screen coordinate systems;
• perform vector, matrix and homogeneous-coordinate calculations;
• apply translation, rotation, scaling, reflection and composite transformations;
• explain orthographic, oblique and perspective projection;
• describe clipping, visibility and the graphics pipeline;
• represent lines and curves parametrically;
• explain the properties and uses of Bézier, B-spline and NURBS curves and surfaces;
• select suitable geometric representations for CAD/CAM applications;
• identify numerical, manufacturing and data-exchange limitations.

1. Coordinate Systems in CAD/CAM

A coordinate system consists of an origin, a set of basis directions and a scale or unit convention. In three-dimensional mechanical design, a point is commonly written as P=(x,y,z), where x, y and z are distances from the coordinate planes. Units must be consistent; a model defined in millimetres must not be interpreted as metres without an explicit conversion.

A right-handed Cartesian system has mutually perpendicular axes and satisfies the right-hand rule: rotating the positive x-axis toward the positive y-axis points in the positive z direction. A left-handed system reverses this orientation. Mixing coordinate handedness can reverse surface normals, mirror models or cause incorrect tool motion.

CAD software typically uses several related systems:

A point’s numerical coordinates have meaning only with respect to its coordinate system. A hole centre at (20, 10, 0) in a part coordinate system is not necessarily at that position in the machine coordinate system.

A cylindrical or spherical feature may be described more naturally with curvilinear coordinates. For cylindrical coordinates, x=r cos θ, y=r sin θ, and z=z, where r is radial distance and θ is the angular coordinate. For spherical coordinates, x=ρ sin φ cos θ, y=ρ sin φ sin θ, and z=ρ cos φ, using the convention that φ is measured from the positive z-axis. Such systems are useful for describing rotational geometry, although Cartesian coordinates remain dominant in transformation algorithms.

Figure: A right-handed three-dimensional Cartesian coordinate system and a point represented by its coordinates.

2. Vectors and Geometric Operations

A point identifies a location, whereas a vector represents a displacement, direction or force. A three-dimensional vector is written v=(v_x,v_y,v_z). Its magnitude is

The formula assumes an orthonormal Cartesian basis. A unit vector in the direction of a nonzero vector is

Vector addition combines displacements component by component. The dot product is

where θ is the angle between nonzero vectors. It is used to calculate angles, projections and visibility. If a·b=0, the vectors are perpendicular.

The cross product is

It produces a vector perpendicular to both a and b, with direction determined by the right-hand rule. Its magnitude is |a×b|=|a||b|sin θ, equal to the area of the parallelogram formed by the vectors. Surface normals in CAD are commonly obtained from the cross product of two surface tangent vectors.

For a line through point P₀ with direction d, the parametric representation is

where t is a real parameter. A line segment between P₀ and P₁ is

The parameter interval is important: unrestricted t describes an infinite line, while 0 ≤ t ≤ 1 describes only the segment.

3. Matrices and Homogeneous Coordinates

Matrices provide a compact way to perform transformations. A three-dimensional point may be represented as a column vector. In ordinary Cartesian coordinates, a 3 × 3 matrix can rotate, scale or reflect a point about the origin. Translation cannot be represented by a 3 × 3 linear matrix because translation does not preserve the origin.

Homogeneous coordinates solve this problem by representing a point as P_h=(x,y,z,1)^T. A direction vector is represented with w = 0: v_h=(v_x,v_y,v_z,0)^T. The distinction prevents translations from changing directions.

A general homogeneous transformation is

where T is a 4 × 4 matrix. Using column vectors, common transformations are:

Translation by (tₓ, tᵧ, t_z):

Scaling by (sₓ, sᵧ, s_z):

Rotation about the z-axis by angle θ:

The angle is measured in the chosen angular unit, normally radians in software. Rotation matrices preserve lengths and angles; uniform scaling preserves angles but changes lengths; nonuniform scaling changes both lengths and generally angles.

Transformation order matters. With column vectors, the rightmost matrix acts first. Thus P′=T R P rotates P first and then translates it. In general, TR≠RT. This is a frequent source of assembly-placement and tool-path errors.

Worked Example 1: Composite Transformation

A point P = (2, 1, 0) is rotated 90° counterclockwise about the z-axis and then translated by (5, −2, 3). Since cos90°=0 and sin90°=1, the rotation gives

P_r=(-1,2,0).

After translation,

P′=(-1+5,2−2,0+3)=(4,0,3).

The result assumes a right-handed system, degrees interpreted only for this calculation, and column-vector transformation order.

For a rigid-body transformation, the upper-left 3 × 3 block is a proper rotation matrix R satisfying R^T R=I and det(R)=+1. The inverse transformation is

where t is the translation vector. This is useful when converting points from world coordinates back into a component’s local coordinates.

4. Projection and the Graphics Pipeline

A graphics system converts geometric data into an image through a sequence commonly called the graphics pipeline:

1. model transformation places local geometry in world coordinates;
2. view transformation expresses the scene relative to the camera;
3. projection transformation maps three-dimensional geometry to a view volume;
4. clipping removes portions outside that volume;
5. perspective division converts homogeneous coordinates to normalized coordinates;
6. viewport transformation maps normalized coordinates to screen pixels;
7. rasterization and visibility processing generate the displayed image.

Orthographic Projection

Orthographic projection uses parallel projectors. Depth does not change apparent size, making it valuable for engineering drawings and dimensioning. A simple projection onto the xy-plane is

More generally, an orthographic view preserves parallelism but may conceal depth. It is unsuitable when realistic visual depth or optical perspective is required.

Oblique Projection

In an oblique projection, projectors remain parallel but are not perpendicular to the projection plane. Cavalier projection may show receding lengths at full scale, whereas cabinet projection commonly shows them at half scale. These conventions are useful for pictorial technical illustrations, but dimensions should normally be taken from orthographic views.

Perspective Projection

Perspective projection models a camera or human observer. With camera at the origin and image plane at distance d, a point (x, y, z), with z > 0, maps approximately to

Here d and z must use the same length units, so x′ and y′ have length units. Objects farther from the camera appear smaller. Actual computer graphics APIs use homogeneous projection matrices and a defined camera convention; signs and depth ranges differ among systems. The chosen convention must therefore be documented rather than assumed.

Clipping

Clipping removes geometry outside a specified viewing or machining region. In two dimensions, a line can be clipped against a rectangular window. In three dimensions, the view volume is bounded by planes such as left, right, top, bottom, near and far. Clipping before rasterization improves efficiency and prevents invalid screen coordinates.

A line segment can be represented parametrically and tested against boundary inequalities. For a rectangular region x_min≤x≤x_max and y_min≤y≤y_max, the algorithm must determine the parameter interval for which all inequalities hold. Algorithms such as Cohen–Sutherland use region codes, while Liang–Barsky uses parametric inequalities. CAD systems also clip section views, drawing sheets, workspaces and simulation domains.

5. Visualization and Geometric Reliability

Visualization is not merely cosmetic. It supports design verification, interference checking, tool accessibility analysis and communication of manufacturing intent. Important methods include wireframe, surface, solid, sectioned, shaded and hidden-line views.

Wireframes show edges but can be ambiguous because occluded edges remain visible. Surface models describe boundaries and support shading, but a surface set may not enclose a valid volume. Solid models represent enclosed volumes and enable mass properties, Boolean operations and collision tests. Boundary representation (B-rep) stores faces, edges and vertices together with topological relationships; constructive solid geometry (CSG) stores a tree of primitive solids and Boolean operations.

Surface normals affect lighting and also carry manufacturing meaning. For a parametric surface S(u,v), tangent vectors are Sᵤ and Sᵥ, and a normal is

provided the tangents are not parallel. Reversing parameter order reverses the normal. Inconsistent normals can produce incorrect shading and, in some CAM systems, incorrect interpretation of inside and outside surfaces.

Numerical visualization must be treated critically. A tessellated display approximates an exact curve or surface with line segments or triangles. Chordal deviation is the maximum distance between the exact geometry and its approximation. If tolerance is too large, a smooth part may appear faceted and a machined surface may violate specification. If tolerance is too small, file size, rendering time and tool-path computation increase.

6. Parametric Geometry

A parametric curve maps a parameter t to a point:

The first derivative C′(t) is tangent to the curve and describes the rate of change with respect to the parameter. If t represents time, its units are velocity; in CAD, t is often dimensionless, so the derivative is a geometric tangent rather than a physical velocity. Curvature for a regular spatial curve is

where C′(t) is nonzero. Curvature measures bending per unit arc length and is important in high-speed machining, vehicle styling and tool-path smoothness.

A parametric surface is

where (u,v) lies in a parameter domain. Surface continuity may be classified as positional (C⁰), tangent (C¹) or curvature (C²). Geometric continuity, written G¹ or G², allows proportional derivatives and is often more relevant to visual smoothness than strict parametric continuity.

7. Bézier Curves and Surfaces

An nth-degree Bézier curve with control points P₀, …, Pₙ is

where the Bernstein basis functions are

and (n choose i)=n!/[i!(n−i)!]. The control points are weighted by basis functions whose sum is one. Consequently, a Bézier curve lies inside the convex hull of its control points and interpolates P₀ and Pₙ. Its endpoint derivatives are

assuming the control points are ordered sequentially.

Bézier curves are intuitive and stable for small, single-span shapes. However, increasing the degree to add design detail affects the entire curve and may cause unwieldy computation or control. Bézier surfaces are formed as a tensor product:

They are useful for patches but complex parts normally require multiple joined patches or a spline representation.

8. B-Spline Curves

A B-spline curve is defined by control points Pᵢ, degree p and a nondecreasing knot vector U = {u₀, u₁, …, uₘ}:

The basis functions are defined recursively. The zeroth-degree basis is

and for p > 0,

A fraction with zero denominator is defined as zero. The valid parameter domain is normally uₚ≤u≤uₘ₋ₚ. B-spline basis functions have local support: changing one control point affects only a limited portion of the curve. This local control is a major advantage over a high-degree Bézier curve.

A clamped knot vector repeats the first and last knots p + 1 times, causing the curve to interpolate its end control points. Internal knot multiplicity affects continuity. For a degree-p spline with internal knot multiplicity r, continuity is generally C^(p−r), assuming regular data and standard spline construction. Thus, repeated knots can create sharper features.

9. NURBS Geometry

NURBS extend B-splines by assigning a positive weight wᵢ to each control point. A NURBS curve is

The denominator must be nonzero over the valid domain. Weights are dimensionless relative values; multiplying every weight by the same positive constant does not change the curve. Rational representation allows exact circles, circular arcs, ellipses and many other conics, whereas ordinary polynomial Bézier and B-spline curves generally represent them only approximately.

NURBS surfaces use two knot vectors, two degrees and a grid of weighted control points:

NURBS are central to industrial CAD because they combine local control, adjustable continuity, compact representation and exact conic geometry. Their limitations include parameterization complexity, sensitivity to invalid weights or knot vectors, and possible interoperability differences between CAD kernels.

10. Worked Example 2: Quadratic Bézier Curve

Let P₀ = (0, 0), P₁ = (2, 4) and P₂ = (5, 0). For a quadratic curve,

C(t)=(1−t)²P₀+2t(1−t)P₁+t²P₂.

At t = 0.5, the basis values are 0.25, 0.5 and 0.25. Therefore,

C(0.5)=0.25(0,0)+0.5(2,4)+0.25(5,0)=(2.25,2.0).

The point lies inside the triangle formed by the control points, illustrating the convex-hull property. The endpoint tangent at t = 0 is C′(0)=2(P₁−P₀)=(4,8).

11. Applied Case: NURBS-Based Mould Design

Background

The following is an illustrative industrial case. A manufacturer must design a plastic injection-mould insert for a consumer-product housing. The housing includes planar mounting regions, blended edges and a circular opening. The insert must be transferred to CAM for three-axis and five-axis machining.

Problem

An early model used many independent polylines and approximate circular curves. The resulting surface displayed small gaps, and the CAM system generated discontinuous tool-path segments near the blended edges. Inspection also revealed that the opening was slightly noncircular after data conversion.

Analysis

The design team identified three causes. First, the wireframe entities did not share exact endpoints, so the boundary was not watertight. Second, approximate polynomial curves represented the circular opening with small geometric error. Third, adjacent surface patches had only positional continuity; their tangent directions changed abruptly, producing an undesirable tool-path transition.

Approach and Solution

The team rebuilt the opening as a rational NURBS curve with appropriate weights, reconstructed the blend as a NURBS surface, and enforced tangent continuity across adjacent patches. Local coordinate systems were used for the insert and then transformed into the assembly and machine datum systems. Model tolerances were specified in millimetres, and the tessellation tolerance was chosen independently from the manufacturing tolerance.

The model was checked for closed topology, consistent surface normals, self-intersections and minimum radii. The CAM programmer generated a roughing operation followed by constant-stepping finishing passes. Verification compared the simulated stock against the nominal NURBS surfaces rather than relying only on the shaded display.

Outcome

The revised model eliminated visible gaps, preserved the exact circular opening during exchange, and produced smoother finishing motion. Inspection showed that the improved result came not from increasing display resolution alone, but from repairing the underlying mathematical and topological representation.

Learning Points

• Geometric accuracy and display quality are different issues.
• Exact rational geometry is valuable when conics or precision surfaces matter.
• Continuity requirements should be established before tool-path generation.
• Coordinate transformations and machine datums must be documented.
• Model validation should include topology, normals, tolerances and independent manufacturing verification.

12. Limitations, Standards and Professional Practice

Mathematical geometry does not by itself guarantee a manufacturable part. A mathematically smooth surface may require an inaccessible tool, excessive axis motion or an impractical setup. Designers must consider cutter radius, stock, fixturing, machine limits, tolerance allocation and inspection strategy.

Data exchange formats can translate geometry differently. STEP-based workflows generally preserve richer solid and surface information than simple polygon formats, while STL represents a tessellated approximation and therefore requires a stated chordal tolerance. A responsible workflow records units, coordinate systems, tolerances, version information and export settings.

Floating-point arithmetic introduces round-off error. Near-coincident points may compare unequal, and Boolean operations may fail when gaps are comparable to numerical tolerance. Robust CAD kernels use tolerance models, geometric predicates and healing operations, but users must not treat automatic healing as proof of design intent.

AI-assisted design and generative systems increasingly produce geometry, optimize forms and suggest tool paths. Their outputs require human verification because an apparently valid mesh may contain non-manifold edges, inaccessible regions, unsuitable wall thickness or untraceable design assumptions. Engineering accountability remains with the responsible organization and professional user.

Best practice is to:

1. define units, axes, datums and tolerances at project initiation;
2. use local coordinate systems for reusable components;
3. document matrix convention and transformation order;
4. choose curve and surface continuity according to function;
5. distinguish exact geometry from visualization tessellation;
6. validate topology, normals, intersections and manufacturability;
7. verify exchanged data against the source model;
8. retain revision history and inspection evidence.

Chapter Summary

CAD and CAM rely on mathematical representations that are precise enough for design, visualization, analysis and manufacturing. Coordinate systems establish the meaning of positions and orientations, while vectors describe directions, displacements, normals and geometric relationships. Matrices, especially homogeneous 4 × 4 matrices, provide a unified method for translation, rotation, scaling and composite placement. Transformation order is significant.

Projection converts three-dimensional scenes into engineering or visual views. Orthographic projection is appropriate for technical documentation, whereas perspective projection conveys depth. Clipping, visibility processing and tessellation form essential parts of the computer-graphics pipeline, but a visually convincing image is not necessarily geometrically valid.

Parametric lines, curves and surfaces support flexible representation and computation. Bézier curves are simple and globally controlled; B-splines provide local control through knot vectors; NURBS add weights and can represent conic sections exactly. Continuity, topology, numerical tolerance and coordinate conventions directly influence CAM reliability. Professional practice therefore combines mathematical understanding with validation, data management and manufacturing judgment.

Key Terms

Coordinate system: An origin, basis directions and scale used to assign coordinates.

Vector: A directed quantity representing displacement, direction or magnitude.

Dot product: A scalar product used for angles, projections and perpendicularity.

Cross product: A three-dimensional vector product producing a normal direction.

Homogeneous coordinates: An augmented coordinate representation that enables translation through matrix multiplication.

Transformation matrix: A matrix that changes the position, orientation or scale of geometric entities.

Rigid-body transformation: A transformation preserving distances and angles.

Projection: The mapping of three-dimensional geometry into a view or image plane.

Clipping: Removal of geometry outside a defined viewing or working region.

Tessellation: Approximation of curves or surfaces by discrete line or polygon elements.

Parametric curve: A curve defined by coordinates that depend on one or more parameters.

Bézier curve: A polynomial curve defined by control points and Bernstein basis functions.

B-spline: A piecewise polynomial spline controlled by degree, knots and control points.

Knot vector: An ordered sequence controlling B-spline parameter intervals and continuity.

NURBS: A weighted rational B-spline representation of curves and surfaces.

Control point: A point influencing, but not necessarily lying on, a curve or surface.

Continuity: The degree to which adjoining geometric segments share position, tangent or curvature behavior.

B-rep: A boundary representation storing faces, edges, vertices and their topology.

Chordal deviation: The distance between exact geometry and its tessellated approximation.

Surface normal: A vector perpendicular to a surface at a specified location.

Review Questions

Short Answer Questions

1. What is the difference between a point and a vector in geometric modelling?
2. Why are homogeneous coordinates used in CAD transformation matrices?
3. State one important difference between orthographic and perspective projection.
4. What role does a knot vector play in a B-spline curve?
5. Why are NURBS especially useful for representing circles and circular arcs?

Long Answer Questions

1. Explain the CAD graphics pipeline from model coordinates to screen coordinates, including transformation, projection, clipping and rasterization.
2. Describe translation, rotation and scaling using homogeneous matrices, and explain why transformation order matters.
3. Compare Bézier, B-spline and NURBS curves in terms of control, continuity, locality and exact geometric representation.
4. Explain how surface normals, topology and tessellation tolerance affect CAD visualization and CAM reliability.
5. Discuss the mathematical and manufacturing considerations involved in selecting a curve or surface representation for a mould insert.

Analytical / Application Questions

1. A point (3, 2, 1) is rotated 90° about the z-axis and then translated by (4, 1, −2). Calculate its final coordinates and explain the transformation order.
2. Design a validation procedure for a NURBS surface exported from one CAD system to another. Include checks for units, topology, continuity, normals and manufacturing tolerance.
3. A CAM system produces a visibly faceted tool path on a nominally smooth surface. Analyse at least three possible mathematical or process causes and recommend corrective actions.

Multiple Choice Questions with Answers

Question 1

Which representation allows translation to be included in a single matrix multiplication with rotation and scaling?

A. A 2 × 2 Cartesian matrix

B. A 3 × 3 Cartesian matrix only

C. A 4 × 4 homogeneous transformation matrix

D. A scalar magnitude

Correct Answer: C. A 4 × 4 homogeneous transformation matrix

Question 2

For vectors a and b, what does a · b = 0 indicate?

A. They are parallel

B. They are perpendicular, assuming both are nonzero

C. They have equal magnitudes

D. They define a unit vector

Correct Answer: B. They are perpendicular, assuming both are nonzero

Question 3

A point (2, 1, 0) is rotated 90° counterclockwise about the z-axis. What is the result?

A. (1, 2, 0)

B. (−1, 2, 0)

C. (2, −1, 0)

D. (−2, −1, 0)

Correct Answer: B. (−1, 2, 0)

Question 4

Which projection preserves parallelism and does not make distant objects appear smaller?

A. Perspective projection

B. Orthographic projection

C. Spherical projection

D. Radial projection

Correct Answer: B. Orthographic projection

Question 5

For a quadratic Bézier curve, how many control points are required?

A. Two

B. Three

C. Four

D. The number is always variable and undefined

Correct Answer: B. Three

Question 6

What is a principal advantage of B-spline local support?

A. Every control point changes the entire curve

B. Changing one control point affects only a limited region

C. The curve cannot have internal knots

D. It removes the need for a parameter

Correct Answer: B. Changing one control point affects only a limited region

Question 7

What additional information do NURBS use compared with ordinary B-splines?

A. Pixel colours

B. Surface normals only

C. Weights for control points

D. Machine spindle speed

Correct Answer: C. Weights for control points

Question 8

For a parametric surface S(u,v), which expression produces a surface normal when the tangents are not parallel?

A. Sᵤ + Sᵥ

B. Sᵤ · Sᵥ

C. Sᵤ × Sᵥ

D. S(u,v) ÷ u

Correct Answer: C. Sᵤ × Sᵥ

Question 9

At t = 0.5, a quadratic Bézier curve with control points (0,0), (2,4), and (5,0) has which point?

A. (1.5, 2)

B. (2, 2)

C. (2.25, 2)

D. (3.5, 4)

Correct Answer: C. (2.25, 2)

Question 10

What does clipping accomplish in a graphics pipeline?

A. It increases the physical size of a model

B. It removes geometry outside a defined viewing or working region

C. It converts every surface into a NURBS surface

D. It guarantees manufacturing accessibility

Correct Answer: B. It removes geometry outside a defined viewing or working region