. . . . . . . . . . . .

 

The datasheet provided does not show the internals. You would have to go through the process of testing 2 or more chips to see if they fight and thus need a diode to isolate their outputs. The LM35 diagram clearly shows that an overvoltage on its output pin will cause the internal amp to shut down and thus leave only the highest sensor outputting.

 

. . . . . . . . . . . .

 

The diode idea is rather dubious, the forward voltage of your average silicon diode is 600-700 mv, your sensor output is just barely enough to turn the diodes on. What you are seeing on the meter is the output actually floating- the diode junction is not in conduction. Adding a resistor to GND will help stabilize the situation a bit, but...

The Diode drop voltage is unpredictable and equal to a large percentage of your signal voltage, when the temperature drops the diode voltage drop will prevail and the output floats. The accuracy you will get will be horrible under the best of circumstances.

What you need is an opamp circuit that buffers the signal from your sensor to a driver that has very low output impedance when sourcing current and high impedance when sinking current, so when the outputs are tied together the highest one wins.

I would also amplify the signal voltage so that cable voltage drops become a smaller portion of the signal, making it more accurate. (use higher supply voltage)

You could add one more wire and make the output signal differential, which would also improve the accuracy.

 

. . . . . . . . . . . . . .