Biomechanical Energy Harvesting: Generating Electricity During Walking with Minimal User Effort

See allHide authors and affiliations

Science  08 Feb 2008:
Vol. 319, Issue 5864, pp. 807-810
DOI: 10.1126/science.1149860


We have developed a biomechanical energy harvester that generates electricity during human walking with little extra effort. Unlike conventional human-powered generators that use positive muscle work, our technology assists muscles in performing negative work, analogous to regenerative braking in hybrid cars, where energy normally dissipated during braking drives a generator instead. The energy harvester mounts at the knee and selectively engages power generation at the end of the swing phase, thus assisting deceleration of the joint. Test subjects walking with one device on each leg produced an average of 5 watts of electricity, which is about 10 times that of shoe-mounted devices. The cost of harvesting—the additional metabolic power required to produce 1 watt of electricity—is less than one-eighth of that for conventional human power generation. Producing substantial electricity with little extra effort makes this method well-suited for charging powered prosthetic limbs and other portable medical devices.

Humans are a rich source of energy. An average-sized person stores as much energy in fat as a 1000-kg battery (1, 2). People use muscle to convert this stored chemical energy into positive mechanical work with peak efficiencies of about 25% (3). This work can be performed at a high rate, with 100 W easily sustainable (1). Many devices take advantage of human power capacity to produce electricity, including hand-crank generators as well as wind-up flashlights, radios, and mobile phone chargers (4). A limitation of these conventional methods is that users must focus their attention on power generation at the expense of other activities, typically resulting in short bouts of generation. For electrical power generation over longer durations, it would be desirable to harvest energy from everyday activities such as walking.

It is a challenge, however, to produce substantial electricity from walking. Most energy-harvesting research has focused on generating electricity from the compression of the shoe sole, with the best devices generating 0.8 W (4). A noteworthy departure is a spring-loaded backpack (5) that harnesses the vertical oscillations of a 38-kg load to generate as much as 7.4 W of electricity during fast walking. This device has a markedly low “cost of harvesting” (COH), a dimensionless quantity defined as the additional metabolic power in watts required to generate 1 W of electrical power Embedded Image(1) where Δ refers to the difference between walking while harvesting energy and walking while carrying the device but without harvesting energy. The COH for conventional power generation is simply related to the efficiency with which (i) the device converts mechanical work to electricity and (ii) muscles convert chemical energy into positive work Embedded Image(2) Embedded Image The backpack's device efficiency is about 31% (5), and muscle's peak efficiency is about 25% (3), yielding an expected COH of 12.9. But the backpack's actual COH of 4.8 ± 3.0 (mean ± SD) is less than 40% of the expected amount. Its economy appears to arise from reducing the energy expenditure of walking with loads (6, 7). No device has yet approached the power generation of the backpack without the need to carry a heavy load.

We propose that a key feature of how humans walk may provide another means of economical energy harvesting. Muscles cyclically perform positive and negative mechanical work within each stride (Fig. 1A) (8). Mechanical work is required to redirect the body's center of mass between steps (9, 10) and simply to move the legs back and forth (11, 12). Even though the average mechanical work performed on the body over an entire stride is zero, walking exacts a metabolic cost because both positive and negative muscle work require metabolic energy (3). Coupling a generator to leg motion would generate electricity throughout each cycle, increasing the load on the muscles during acceleration but assisting them during deceleration (Fig. 1B). Although generating electricity during the acceleration phase would exact a substantial metabolic cost, doing so during the deceleration phase would not, resulting in a lower COH than for conventional generation. An even lower COH could be achieved by selectively engaging the generator only during deceleration (Fig. 1C), similar to how regenerative braking generates power while decelerating a hybrid car (13). Here, “generative braking” produces electricity without requiring additional positive muscle power (14). If implemented effectively, metabolic cost could be about the same as that for normal walking, so energy would be harvested with no extra user effort (15).

Fig. 1.

Theoretical advantages of generative braking during cyclic motion, comparing the back-and-forth motion of the knee joint without power generation (A) against a generator operating continuously (B) and against a generator operating only during braking (C). Each column of plots shows the rate of work performed by muscles (work rate) and the electricity (elect. power) generated over time, as well as the average metabolic power expended by the human and the resulting average electrical power (ave. power bar graphs). In (B) and (C), work rate is compared against that for (A), denoted by dashed lines, and average power is shown as the difference (Δ ave. power) with respect to (A). COH is defined as the ratio of the electrical to metabolic Δ ave. powers.

We developed a wearable, knee-mounted prototype energy harvester to test the generative-braking concept (Fig. 2). Although other joints might suffice, we focused on the knee because it performs mostly negative work during walking (16). The harvester comprises an orthopedic knee brace configured so that knee motion drives a gear train through a one-way clutch, transmitting only knee extension motion at speeds suitable for a dc brushless motor that serves as the generator (17). For convenient testing, generated electrical power is then dissipated with a load resistor rather than being used to charge a battery. The device efficiency, defined as the ratio of the electrical power output to the mechanical power input, was empirically estimated to be no greater than 63%, yielding an estimated COH for conventional generation of 6.4 (Eq. 2). A potentiometer senses knee angle, which is fed back to a computer controlling a relay switch in series with the load resistor, allowing the electrical load to be selectively disconnected in real time. For generative braking, we programmed the harvester to engage only during the end of the swing phase (Fig. 3), producing electrical power while simultaneously assisting the knee flexor muscles in decelerating the knee. We compared this mode against a continuous-generation mode that harvests energy whenever the knee is extending (18). We could also manually disengage the clutch and completely decouple the gear train and generator from knee motion. This disengaged mode served as a control condition to estimate the metabolic cost of carrying the harvester mass, independent of the cost of generating electricity.

Fig. 2.

Biomechanical energy harvester. (A) The device has an aluminum chassis (green) and generator (blue) mounted on a customized orthopedic knee brace (red), totaling 1.6-kg mass, with one worn on each leg. (B) The chassis contains a gear train that converts low velocity and high torque at the knee into high velocity and low torque for the generator, with a one-way roller clutch that allows for selective engagement of the gear train during knee extension only and no engagement during knee flexion. (C) The schematic diagram shows how a computer-controlled feedback system determines when to generate power using knee-angle feedback, measured with a potentiometer mounted on the input shaft. Generated power is dissipated in resistors. Rg, generator internal resistance; RL, output load resistance; E(t), generated voltage.

Fig. 3.

Timing of power generation during walking. Time within a stride cycle, beginning with the swing phase, is shown at the bottom. The shaded bars indicate when the knee is extending and the energy harvester's clutch is engaged. (A) The pattern of knee mechanical power during normal walking illustrates that the knee typically generates a large amount of negative power at the end of the swing phase (16). (B) Mechanical power performed on the harvester over time, shown for continuous generation (red line) and generative braking (blue line). (C) Generated electrical power over time, also for both types of generation.

Energy-harvesting performance was tested on six male subjects who wore a device on each leg while walking on a treadmill at 1.5 m s–1. We estimated metabolic cost using a standard respirometry system and measured the electrical power output of the generator (Fig. 3C). In the continuous-generation mode (Fig. 4A), subjects generated 7.0 ± 0.7 W of electricity with an insignificant 18 ± 24 W (P = 0.07) increase in metabolic cost over that of the control condition (19). In the generative-braking mode (Fig. 4B), subjects generated 4.8 ± 0.8 W of electricity with an insignificant 5 ± 21 W increase in metabolic cost as compared with that of the control condition (P = 0.6). For context, this electricity is sufficient to power 10 typical cell phones simultaneously (5). The results demonstrate that substantial electricity could be generated with minimal increase in user effort.

Fig. 4.

Average metabolic cost and generated electricity for continuous generation (A) and generative braking (B), with change in metabolic cost (Δ average power) shown relative to the control condition. (C) COH (see Fig. 1) for continuous generation and generative braking as compared against that for conventional generation (dashed line). In both modes, a fraction of the harvested energy is generated from the deceleration of the knee rather than directly from muscle action. Error bars in (A) to (C) indicate SD. Asterisks indicate significant differences with conventional generation (*) and between continuous generation and generative braking (**) (P < 0.05 for all comparisons).

The corresponding COH values highlight the advantage of generative braking (Fig. 4). Average COH in generative braking was only 0.7 ± 4.4; less than 1 W of metabolic power was required to generate 1 W of electricity. This is significantly less than the COH of 6.4 expected for conventional generation (P = 0.01). The COH in continuous generation, 2.3 ± 3.0, was also significantly lower than that for conventional generation (P = 0.01), indicating that the former mode also generated some of its electricity from the deceleration of the knee. The difference between the two modes, 2.2 ± 0.7 W of electricity, came at a difference in metabolic cost of 13 ± 12 W (P = 0.05). A COH taken from the average ratio of these differences yields 5.7 ± 6.2, which is nearly the same as that expected of conventional generation (P = 0.4). This indicates that continuous generation of power at the knee during walking produces electricity partially by conventional generation with a high COH and partially by generative braking with a very low COH. But generative braking, with less than one-eighth the COH of conventional generation, benefits almost entirely from the deceleration of the knee.

This preliminary demonstration could be improved substantially. We constructed the prototype for convenient experimentation, leading to a control condition about 20% more metabolically costly than normal walking: The disengaged-clutch mode required an average metabolic power of 366 ± 63 W as compared with 307 ± 64 W for walking without wearing the devices. The increase in cost is due mainly to the additional mass and its location, because the lower a given mass is placed, the more expensive it is to carry (20, 21). Although the current increase in metabolic cost is unacceptably high for most practical implementations, revisions to improve the fit, weight, and efficiency of the device can not only reduce the cost but also increase the generated electricity. A generator designed specifically for this application could have lower internal losses and require a smaller, lighter gear train. Commercially available gear trains can have much lower friction and higher efficiency, in more compact and lightweight forms. Relocating the device components higher would decrease the metabolic cost of carrying that mass. A more refined device would also benefit from a more form-fitting knee brace made out of a more lightweight material such as carbon fiber.

Several potential applications are especially suited for generative braking. These include lighting and communications needs for the quarter of the world's population who currently live without electricity supply (22). Innovative prosthetic knees and ankles use motors to assist walking, but battery technology limits their power and working time (2325). Energy harvesters worn on human joints may prove useful for powering the robotic artificial joints. In implantable devices, such as neurostimulators and drug pumps, battery power limits device sophistication, and battery replacement requires surgery (26). A future energy harvester might be implanted alongside such a device, perhaps in parallel with a muscle, and use generative braking to provide substantial power indefinitely. Generative braking might then find practical applications in forms very different from that demonstrated here.

Supporting Online Material

Materials and Methods

Figs. S1 to S3

Table S1


Movies S1 to S4

References and Notes

View Abstract

Stay Connected to Science

Navigate This Article