Having thought several times about purchasing a flashlight that could be recharged simply by shaking, i.e. I'd be the one to supply the energy once the original purchase was made, I found this article interesting. It comes from the made by monkeys site:

http://www.electronicsweekly.com/blogs/engineering-design-problems/2007/06/flashlight-has-dismal-light-ou-1.html 

Those portable shake-to-charge devices seem like a good idea, especially for the battery challenged, but how well does this sort of thing really work? Engineer Dave Johnson took this flashlight to the test and found the performance decidedly lacking:

"The maker used some very cheap 1N4001 diodes in the bridge rectifier circuit instead of more efficient Schottky diodes. They also used a small 0.5 Farad cap with a 5.5V rating. I noticed that that this kind of super capacitor was originally designed for maintaining data in memory chips and has a rather high internal equivalent series resistance. This reduces the overall efficiency, since the device can't be charged or discharged very quickly. Some of the power that should go to the LED will end up being dissipated inside the capacitor. Better super capacitors do exist.Most white LEDs draw about 20ma of current with a voltage of about 3.6 volts. As the voltage drops from 3.6v, the current will also be lower. Without any regulation, the circuit will not have a consistent light output. I measured the LED current in this circuit at only a few milliamps, even after many minutes of vigorous shaking. This suggests that they decided to sacrifice light intensity for light duration.
View circuit

The human mechanical power to electrical power conversion efficiency for a shaking device, such as this flashlight, is poor. To measure how much power I could get from the shaking magnet generator, I first completed the flashlight dissection process by disconnecting the coil from the flashlight circuit. I then connected the coil to a Schottky diode bridge, made from four 1N5817 diodes. These diodes have a much lower 0.35V drop instead of the 1.0V for the 1N4001 diode. I then placed a high quality 10 Farad super capacitor from Maxwell (www.maxwell.com) across the output of the bridge.

To measure the capacitor voltage, I connected a digital voltmeter across the cap. Before the test, I made sure the capacitor was completely discharged. I started a stopwatch, and then started shaking the flashlight's magnet. After 120 seconds, the capacitor was charged up to 1.0 volts. This corresponds to an energy increase of 5 joules using the equation 0.5CVV, where C is 10 Farads and V is the 1.0 volts across the capacitor. So, I got 5 joules (watt-seconds) of energy in 120 seconds. That means that the magnet shaking was only able to produce about 0.042 watts of power. This is a dismal amount -- and I am no weakling! To put this into perspective, a single 1.5" x 1.5 " solar cell, placed in bright sunlight, would generate more power than the shaking magnet generator. I bet many of the hand crank generator flashlights I have seen for sale would do much better. A pull string type generator would work even better.

But let's imagine a different kind of flashlight altogether.

I like the idea of using a pull string human motion over crank. I think this technique would be a much more efficient way to convert human power to electrical power. This device would be a miniature version of the technique often used to start a lawn mower. The imaginary device would be held in one hand, while the other arm would pull the string. Each pull of the string would spin a flywheel up to a high RPM, which would be connected to a brushless motor, acting as alternator. I imagine that such a device could generate several watts of power, perhaps as much as 20 watts. Such a machine would produce 50 or 100 times more power than the shaking magnet technique.

Let's then connect the output of the three phase alternator to a three phase bridge rectifier. The output of the bridge would then be connected to a bank of quality super capacitors rated at 2.5 volts. Perhaps the total capacitance would be about 50 Farads. I would then use an efficient DC-to-DC converter (Maybe using Texas Instrument's TPS61070 device), designed to maintain a constant current flow to one or more LEDs. One LED might be adequate but an array of 7 LEDs would be even better.

Let's say the pull string approach could generate 5 watts of power. Then, a 50 Farad cap would be charged up to 2.5 volts in less than a minute. Once charged, a DC to DC converter with an 80% efficiency would keep a single 20ma white LED going for 30 minutes. Such a flashlight would be much brighter than the above device. A larger 350 farad capacitor from www.maxwell.com might be used to power a 7 LED flashlight for 30 minutes after 7 minutes of charging. If a 20 watt device could be developed, it might be powerful enough to charge up a dead automotive lead acid battery after a few minutes of string pulling to start a car. Such a device might be small enough to fit into a car's glove box and could make a nice Christmas gift.