The first bearings were wood, but ceramic or glass can be used, and steel is quite common. The path that a bearing travels is called a "race". Bearings need to be as hard as the race that they are rolling against, or either the bearing or race will wear out prematurely. Different types exist for both linear and rotary motion.
Linear bearings
An early type of linear bearing was an arrangement of tree trunks laid down under sleds. This technology is known to date at least as far back as the construction of the Pyramids of Giza. Modern linear bearings use a similar principle, sometimes with balls in place of rollers.
Rotary bearings
Rotary bearings are required for many applications, from heavy-duty use in vehicle axles and machine shafts, to precision clock parts. The simplest rotary bearing is the sleeve bearing, which is just a cylinder inserted between the wheel and its axle. This was followed by the roller bearing, in which the sleeve was replaced by a number of cylindrical rollers. Each roller behaves as an individual wheel. The first practical caged-roller bearing was invented by horologist John Harrison in his H3 chronometer of 1760.
Ball Bearings
The ball bearing is a variation of the roller bearing. It is easier (cheaper) to produce balls of high accuracy than rollers of high accuracy; ball bearings can support both radial and axial loads; ball bearings offer very low friction; and ball bearings can operate when the load is somewhat misaligned. Roller bearings are still used, however, as rollers distribute the load over a greater area within the bearing and thus a given size of roller bearing can carry a higher load.
An early example of a wooden ball bearing, supporting a rotating table, was retrieved from the remains of a Roman ship in Lake Nemi, Italy. The wreck was dated to 40 BC. Leonardo da Vinci is said to have described a type of ball bearing around the year 1500. One of the issues with ball bearings is that they can rub against each other, causing additional friction, but this can be prevented by enclosing the balls in a cage. The captured, or caged, ball bearing was originally described by Galileo in the 1600s. The mounting of bearings into a set was not accomplished for many years after that. The first patent for a ball race was by Philip Vaughan of Carmarthen in 1794. The modern, self-aligning design of ball bearing is attributed to Sven Wingquist of the SKF ball-bearing manufacturer in 1907.
Each bearing race is a ring with a groove or face where the balls rest. The groove is usually shaped so the ball is a slightly loose fit in the groove. Thus, in principle, the ball contacts each race at a single point. However, a load on an infinitely small point would cause infinitely high contact pressure. In practice, the ball deforms (flattens) slightly where it contacts each race, much as a tire flattens where it touches the road. The race also dents slightly where each ball presses on it. Thus, the contact between ball and race is of finite size and has finite pressure. Note, however, that the deformed ball and race do not roll entirely smoothly, because different parts of the ball are moving at different speeds as it rolls. Thus, there are opposing forces and sliding motions at each ball/race contact. Overall, these cause bearing drag and friction.
Rolling-Element Bearing Design
There are now many types of roller bearings, each tuned for a specific kind of load and with specific advantages and disadvantages. For example:
There are two usual limits to the lifetime or load capacity of a bearing:
fatigue and pressure-induced welding. Fatigue is when a material breaks after it is repeatedly bent and released. Where the ball or roller touches the race there is always some bending, and hence a risk of fatigue. Smaller balls or rollers bend more sharply, and so tend to fatigue faster. Pressure-induced welding is when two metal pieces are pressed together at very high pressure and they become one. Although balls, rollers and races may look smooth, they are microscopically rough. Thus, there are high-pressure spots which push away the bearing lubricant. Sometimes, the resulting metal-to-metal contact welds a tiny part of the ball or roller to the race. As the bearing continues to rotate, the weld is then torn apart, but it may leave race welded to bearing or bearing welded to race.
Although bearings tend to wear out with use, designers can make tradeoffs of bearing size and cost versus lifetime. A bearing can have essentially infinite lifetime -- longer than the rest of the machine -- if it is kept cool, clean, lubricated, is run within the rated load, and if the bearing materials are sufficiently free of microscopic defects. Note that cooling, lubrication, and sealing are thus important parts of the bearing design.
Rolling-element bearings often work well in non-ideal conditions. But sometimes minor problems cause bearings to fail quickly and mysteriously. For example, with a stationary (non-rotating) load, small vibrations can gradually press out the lubricant between the races and rollers or balls. Without lubricant the bearing fails, even though it is not rotating and thus is apparently not being used! For these sorts of reasons, much of bearing design is about failure analysis.
Some modern bearing assemblies require routine addition of lubricants, while others are factory sealed, requiring no further maintenance for the life of the mechanical assembly. The lubricant is intended to reduce friction. However, if the lubricant becomes contaminated by hard particles, such as steel chips from the race or bearing, sand, or grit, the lubricant quickly begins to act as a grinding compound. This greatly reduces the operating life of the bearing assembly.
A standard reference on bearing design is ``Rolling Bearing Analysis'', by Tedric A. Harris, ISBN 0801988713.