Brakes completely repeatedly failing at high temperature

I will have a look. It should actually be the other way when the motor gets warm - braking possible, but no acceleration.

Could it be that this is related to inverting the motor direction? I have the impression that the "invert motor direction" tickmark has a few bugs. Like in this case, braking would be a positive current and acellerating a negative one. If those are not properly handled compared to the invert motor direction then that would lead to this effect. 

Not yet, but I will have a look this weekend. If you have invert motor direction activated, can you try deactivating it, switching two phases and see if it makes any difference?

Paul15578 is right. Problem is that when you come into the range where the amps should be reduced by the temperature (not cut off) then only the higher motor max or min will be reduced. The other value will be 0. So when your motor max is higher then motor min the brakes will be shut off.

// Temperature MOSFET
	float lo_max_mos = 0.0;
	float lo_min_mos = 0.0;
	if (m_temp_fet < conf->l_temp_fet_start) {
		lo_min_mos = conf->l_current_min;
		lo_max_mos = conf->l_current_max;
	} else if (m_temp_fet > conf->l_temp_fet_end) {
		lo_min_mos = 0.0;
		lo_max_mos = 0.0;
		mc_interface_fault_stop(FAULT_CODE_OVER_TEMP_FET);
	} else {
		float maxc = fabsf(conf->l_current_max);
		if (fabsf(conf->l_current_min) > maxc) {
			maxc = fabsf(conf->l_current_min);
		}

		maxc = utils_map(m_temp_fet, conf->l_temp_fet_start, conf->l_temp_fet_end, maxc, 0.0);

		if (fabsf(conf->l_current_max) > maxc) {
			lo_max_mos = SIGN(conf->l_current_max) * maxc;
		}

		if (fabsf(conf->l_current_min) > maxc) {
			lo_min_mos = SIGN(conf->l_current_min) * maxc;
		}
	}

Problem can be solved by reducing both values in mosfet and motor temperature

	// Temperature MOSFET
	float lo_max_mos = 0.0;
	float lo_min_mos = 0.0;
	if (m_temp_fet < conf->l_temp_fet_start) {
		lo_min_mos = conf->l_current_min;
		lo_max_mos = conf->l_current_max;
	} else if (m_temp_fet > conf->l_temp_fet_end) {
		lo_min_mos = 0.0;
		lo_max_mos = 0.0;
		mc_interface_fault_stop(FAULT_CODE_OVER_TEMP_FET);
	} else {
		lo_min_mos = SIGN(conf->l_current_min) * utils_map(m_temp_fet, conf->l_temp_fet_start, conf->l_temp_fet_end, fabsf(conf->l_current_min), 0.0);
		lo_max_mos = SIGN(conf->l_current_max) * utils_map(m_temp_fet, conf->l_temp_fet_start, conf->l_temp_fet_end, fabsf(conf->l_current_max), 0.0);
            
		/*float maxc = fabsf(conf->l_current_max);
		if (fabsf(conf->l_current_min) > maxc) {
			maxc = fabsf(conf->l_current_min);
		}

		maxc = utils_map(m_temp_fet, conf->l_temp_fet_start, conf->l_temp_fet_end, maxc, 0.0);

		if (fabsf(conf->l_current_max) > maxc) {
			lo_max_mos = SIGN(conf->l_current_max) * maxc;
		}

		if (fabsf(conf->l_current_min) > maxc) {
			lo_min_mos = SIGN(conf->l_current_min) * maxc;
		}*/
	}

	// Temperature MOTOR
	float lo_max_mot = 0.0;
	float lo_min_mot = 0.0;
	if (m_temp_motor < conf->l_temp_motor_start) {
		lo_min_mot = conf->l_current_min;
		lo_max_mot = conf->l_current_max;
	} else if (m_temp_motor > conf->l_temp_motor_end) {
		lo_min_mot = 0.0;
		lo_max_mot = 0.0;
		mc_interface_fault_stop(FAULT_CODE_OVER_TEMP_MOTOR);
	} else {
		lo_min_mot = SIGN(conf->l_current_min) * utils_map(m_temp_motor, conf->l_temp_motor_start, conf->l_temp_motor_end, fabsf(conf->l_current_min), 0.0);
		lo_max_mot = SIGN(conf->l_current_max) * utils_map(m_temp_motor, conf->l_temp_motor_start, conf->l_temp_motor_end, fabsf(conf->l_current_max), 0.0);
		/*    
		float maxc = fabsf(conf->l_current_max);
		if (fabsf(conf->l_current_min) > maxc) {
			maxc = fabsf(conf->l_current_min);
		}

		maxc = utils_map(m_temp_motor, conf->l_temp_motor_start, conf->l_temp_motor_end, maxc, 0.0);

		if (fabsf(conf->l_current_max) > maxc) {
			lo_max_mot = SIGN(conf->l_current_max) * maxc;
		}

		if (fabsf(conf->l_current_min) > maxc) {
			lo_min_mot = SIGN(conf->l_current_min) * maxc;
		}*/
	}

Well spotted @Ackmaniac, thanks! Such an obvious mistake. The reason I never noticed it is that I have always used the same acceleration and braking current, which would not trigger the bug. The more difference there is between the acceleration and braking currents, the wider the temperature region where the lower of the two would be set to 0. If this is the only problem, the bug is independent of whether the motor direction is inverted or not.

I went with the following solution:

// Temperature MOSFET
	float lo_min_mos = conf->l_current_min;
	float lo_max_mos = conf->l_current_max;
	if (m_temp_fet < conf->l_temp_fet_start) {
		// Keep values
	} else if (m_temp_fet > conf->l_temp_fet_end) {
		lo_min_mos = 0.0;
		lo_max_mos = 0.0;
		mc_interface_fault_stop(FAULT_CODE_OVER_TEMP_FET);
	} else {
		float maxc = fabsf(conf->l_current_max);
		if (fabsf(conf->l_current_min) > maxc) {
			maxc = fabsf(conf->l_current_min);
		}

		maxc = utils_map(m_temp_fet, conf->l_temp_fet_start, conf->l_temp_fet_end, maxc, 0.0);

		if (fabsf(conf->l_current_min) > maxc) {
			lo_min_mos = SIGN(conf->l_current_min) * maxc;
		}

		if (fabsf(conf->l_current_max) > maxc) {
			lo_max_mos = SIGN(conf->l_current_max) * maxc;
		}
	}

	// Temperature MOTOR
	float lo_min_mot = conf->l_current_min;
	float lo_max_mot = conf->l_current_max;
	if (m_temp_motor < conf->l_temp_motor_start) {
		// Keep values
	} else if (m_temp_motor > conf->l_temp_motor_end) {
		lo_min_mot = 0.0;
		lo_max_mot = 0.0;
		mc_interface_fault_stop(FAULT_CODE_OVER_TEMP_MOTOR);
	} else {
		float maxc = fabsf(conf->l_current_max);
		if (fabsf(conf->l_current_min) > maxc) {
			maxc = fabsf(conf->l_current_min);
		}

		maxc = utils_map(m_temp_motor, conf->l_temp_motor_start, conf->l_temp_motor_end, maxc, 0.0);

		if (fabsf(conf->l_current_min) > maxc) {
			lo_min_mot = SIGN(conf->l_current_min) * maxc;
		}

		if (fabsf(conf->l_current_max) > maxc) {
			lo_max_mot = SIGN(conf->l_current_max) * maxc;
		}
	}

The reason it is implemented in this was and not how Ackmaniac suggested is that only the highest current magnitude should be decreased, as the amount of heating is not direction dependent. Arguments can be made for the other way as well, but I lean a bit more towards only decreasing the highest magnitude.

I will test this a bit as well as the impact of the direction inversion, and update the firmware and vesc tool.

It seems to be working now. You can update to the latest firmware and vesc tool, where it should be fixed. The build script is running now, the downloads should be updated any minute.