|
@@ -36,8 +36,11 @@
|
|
|
#define Z_PLUS 0
|
|
|
#define Z_MINUS 1
|
|
|
|
|
|
-/// 10000 = 1 mm/s
|
|
|
+/// Max. jerk in PrusaSlicer, 10000 = 1 mm/s
|
|
|
#define MAX_DELAY 10000
|
|
|
+#define MIN_SPEED (0.01f / (MAX_DELAY * 0.000001f))
|
|
|
+/// 200 = 50 mm/s
|
|
|
+#define Z_MIN_DELAY 200
|
|
|
#define Z_ACCEL 300
|
|
|
#define XY_ACCEL 1000
|
|
|
|
|
@@ -76,6 +79,11 @@
|
|
|
__typeof__ (max) max_ = (max); \
|
|
|
( a_ < min_ ? min_ : (a_ <= max_ ? a_ : max_)); })
|
|
|
|
|
|
+/// \returns square of the value
|
|
|
+#define SQR(a) \
|
|
|
+ ({ __typeof__ (a) a_ = (a); \
|
|
|
+ (a_ * a_); })
|
|
|
+
|
|
|
/// position types
|
|
|
typedef int16_t pos_i16_t;
|
|
|
typedef long pos_i32_t;
|
|
@@ -365,9 +373,11 @@ int8_t xyzcal_meassure_pinda_hysterezis(int16_t min_z, int16_t max_z, uint16_t d
|
|
|
}
|
|
|
#endif //XYZCAL_MEASSURE_PINDA_HYSTEREZIS
|
|
|
|
|
|
+/// Accelerate up to max.speed (defined by @min_delay_us)
|
|
|
void accelerate(uint8_t axis, int16_t acc, uint16_t &delay_us, uint16_t min_delay_us){
|
|
|
sm4_do_step(axis);
|
|
|
|
|
|
+ /// keep max speed (avoid extra computation)
|
|
|
if (acc > 0 && delay_us == min_delay_us){
|
|
|
delayMicroseconds(delay_us);
|
|
|
return;
|
|
@@ -375,47 +385,66 @@ void accelerate(uint8_t axis, int16_t acc, uint16_t &delay_us, uint16_t min_dela
|
|
|
|
|
|
// v1 = v0 + a * t
|
|
|
// 0.01 = length of a step
|
|
|
- const float d0 = delay_us * 0.000001f;
|
|
|
- const float v1 = (0.01f / d0 + acc * d0);
|
|
|
- uint16_t d1;
|
|
|
- if (v1 <= 0.1f){
|
|
|
- d1 = MAX_DELAY; ///< already too slow so it wants to move back
|
|
|
+ const float t0 = delay_us * 0.000001f;
|
|
|
+ const float v1 = (0.01f / t0 + acc * t0);
|
|
|
+ uint16_t t1;
|
|
|
+ if (v1 <= 0.16f){ ///< slowest speed convertible to uint16_t delay
|
|
|
+ t1 = MAX_DELAY; ///< already too slow so it wants to move back
|
|
|
} else {
|
|
|
- d1 = MAX(min_delay_us, round_to_u16(0.01f / v1));
|
|
|
+ /// don't exceed max.speed
|
|
|
+ t1 = MAX(min_delay_us, round_to_u16(0.01f / v1 * 1000000.f));
|
|
|
}
|
|
|
|
|
|
/// make sure delay has changed a bit at least
|
|
|
- if (d1 == delay_us && acc != 0){
|
|
|
+ if (t1 == delay_us && acc != 0){
|
|
|
if (acc > 0)
|
|
|
- d1--;
|
|
|
+ t1--;
|
|
|
else
|
|
|
- d1++;
|
|
|
+ t1++;
|
|
|
}
|
|
|
|
|
|
- delayMicroseconds(d1);
|
|
|
- delay_us = d1;
|
|
|
-}
|
|
|
-
|
|
|
+ //DBG(_n("%d "), t1);
|
|
|
|
|
|
-uint8_t slow_down_z(uint8_t axis, uint16_t delay_us){
|
|
|
- sm4_do_step(axis);
|
|
|
- delayMicroseconds(delay_us / 3 * 4);
|
|
|
- sm4_do_step(Z_AXIS_MASK);
|
|
|
- delayMicroseconds(delay_us * 2);
|
|
|
- sm4_do_step(Z_AXIS_MASK);
|
|
|
- delayMicroseconds(delay_us * 4);
|
|
|
- return 3;
|
|
|
+ delayMicroseconds(t1);
|
|
|
+ delay_us = t1;
|
|
|
}
|
|
|
|
|
|
-uint8_t speed_up_z(uint8_t axis, uint16_t delay_us){
|
|
|
- sm4_do_step(Z_AXIS_MASK);
|
|
|
- delayMicroseconds(delay_us * 4);
|
|
|
- sm4_do_step(Z_AXIS_MASK);
|
|
|
- delayMicroseconds(delay_us * 2);
|
|
|
- sm4_do_step(Z_AXIS_MASK);
|
|
|
- delayMicroseconds(delay_us / 3 * 4);
|
|
|
- return 3;
|
|
|
-}
|
|
|
+void go_and_stop(uint8_t axis, int16_t dec, uint16_t &delay_us, uint16_t &steps){
|
|
|
+ if (steps <= 0 || dec <= 0)
|
|
|
+ return;
|
|
|
+
|
|
|
+ /// deceleration distance in steps, s = 1/2 v^2 / a
|
|
|
+ uint16_t s = round_to_u16(100 * 0.5f * SQR(0.01f) / (SQR((float)delay_us) * dec));
|
|
|
+ if (steps > s){
|
|
|
+ /// go steady
|
|
|
+ sm4_do_step(axis)
|
|
|
+ delayMicroseconds(delay_us);
|
|
|
+ } else {
|
|
|
+ /// decelerate
|
|
|
+ accelerate(axis, -dec, &delay_us, min_delay_us);
|
|
|
+ }
|
|
|
+ --steps;
|
|
|
+}
|
|
|
+
|
|
|
+// uint8_t slow_down_z(uint8_t axis, uint16_t delay_us){
|
|
|
+// sm4_do_step(axis);
|
|
|
+// delayMicroseconds(delay_us / 3 * 4);
|
|
|
+// sm4_do_step(Z_AXIS_MASK);
|
|
|
+// delayMicroseconds(delay_us * 2);
|
|
|
+// sm4_do_step(Z_AXIS_MASK);
|
|
|
+// delayMicroseconds(delay_us * 4);
|
|
|
+// return 3;
|
|
|
+// }
|
|
|
+
|
|
|
+// uint8_t speed_up_z(uint8_t axis, uint16_t delay_us){
|
|
|
+// sm4_do_step(Z_AXIS_MASK);
|
|
|
+// delayMicroseconds(delay_us * 4);
|
|
|
+// sm4_do_step(Z_AXIS_MASK);
|
|
|
+// delayMicroseconds(delay_us * 2);
|
|
|
+// sm4_do_step(Z_AXIS_MASK);
|
|
|
+// delayMicroseconds(delay_us / 3 * 4);
|
|
|
+// return 3;
|
|
|
+// }
|
|
|
|
|
|
void xyzcal_scan_pixels_32x32_Zhop(int16_t cx, int16_t cy, int16_t min_z, int16_t max_z, uint16_t delay_us, uint8_t* pixels){
|
|
|
if(!pixels)
|
|
@@ -423,8 +452,12 @@ void xyzcal_scan_pixels_32x32_Zhop(int16_t cx, int16_t cy, int16_t min_z, int16_
|
|
|
int16_t z = _Z;
|
|
|
int16_t z_trig;
|
|
|
uint16_t line_buffer[32];
|
|
|
- uint16_t current_delay_us = MAX_DELAY;
|
|
|
+ uint16_t current_delay_us = MAX_DELAY; ///< defines current speed
|
|
|
xyzcal_lineXYZ_to(cx - 1024, cy - 1024, min_z, delay_us, 0);
|
|
|
+ uint16_t min_decel_z;
|
|
|
+ int16_t start_z;
|
|
|
+ int16_t last_top_z;
|
|
|
+
|
|
|
for (uint8_t r = 0; r < 32; r++){ ///< Y axis
|
|
|
xyzcal_lineXYZ_to(_X, cy - 1024 + r * 64, z, delay_us, 0);
|
|
|
for (int8_t d = 0; d < 2; ++d){ ///< direction
|
|
@@ -433,40 +466,58 @@ void xyzcal_scan_pixels_32x32_Zhop(int16_t cx, int16_t cy, int16_t min_z, int16_
|
|
|
z = _Z;
|
|
|
sm4_set_dir(X_AXIS, d);
|
|
|
for (uint8_t c = 0; c < 32; c++){ ///< X axis
|
|
|
-
|
|
|
+
|
|
|
/// move up to un-trigger (surpress hysteresis)
|
|
|
sm4_set_dir(Z_AXIS, Z_PLUS);
|
|
|
+ /// speed up from stop, go half the way
|
|
|
current_delay_us = MAX_DELAY;
|
|
|
- while (z < max_z && _PINDA){
|
|
|
- accelerate(Z_AXIS_MASK, Z_ACCEL, current_delay_us, delay_us);
|
|
|
- z++;
|
|
|
+ for (start_z = z; z < (max_z + start_z) / 2; ++z){
|
|
|
+ if (!_PINDA){
|
|
|
+ last_top_z = z;
|
|
|
+ break;
|
|
|
+ }
|
|
|
+ accelerate(Z_AXIS_MASK, Z_ACCEL, current_delay_us, Z_MIN_DELAY);
|
|
|
+ }
|
|
|
+
|
|
|
+ if(_PINDA){
|
|
|
+ uint16_t steps_to_go = MAX(0, max_z - z);
|
|
|
+ while (_PINDA && z < max_z){
|
|
|
+ go_and_stop(Z_AXIS_MASK, Z_ACCEL, current_delay_us, steps_to_go);
|
|
|
+ ++z;
|
|
|
+ }
|
|
|
+ last_top_z = z;
|
|
|
+ }
|
|
|
+ /// slow down to stop
|
|
|
+ while (current_delay_us < MAX_DELAY){
|
|
|
+ accelerate(Z_AXIS_MASK, -Z_ACCEL, current_delay_us, Z_MIN_DELAY);
|
|
|
+ ++z;
|
|
|
}
|
|
|
- int16_t last_top_z = z;
|
|
|
- z += slow_down_z(delay_us);
|
|
|
|
|
|
/// move down to trigger
|
|
|
sm4_set_dir(Z_AXIS, Z_MINUS);
|
|
|
-
|
|
|
/// speed up
|
|
|
- do (){
|
|
|
- if (z <= min_z || _PINDA) break;
|
|
|
- sm4_do_step(Z_AXIS_MASK);
|
|
|
- delayMicroseconds(delay_us * 4);
|
|
|
- if (z <= min_z || _PINDA) break;
|
|
|
- sm4_do_step(Z_AXIS_MASK);
|
|
|
- delayMicroseconds(delay_us * 2);
|
|
|
- if (z <= min_z || _PINDA) break;
|
|
|
- sm4_do_step(Z_AXIS_MASK);
|
|
|
- delayMicroseconds(delay_us / 3 * 4);
|
|
|
- } while (0);
|
|
|
-
|
|
|
- while (z > min_z && !_PINDA){
|
|
|
- sm4_do_step(Z_AXIS_MASK);
|
|
|
- delayMicroseconds(delay_us);
|
|
|
- z--;
|
|
|
+ current_delay_us = MAX_DELAY;
|
|
|
+ for (start_z = z; z > (min_z + start_z) / 2; --z){
|
|
|
+ if (_PINDA){
|
|
|
+ z_trig = z;
|
|
|
+ break;
|
|
|
+ }
|
|
|
+ accelerate(Z_AXIS_MASK, Z_ACCEL, current_delay_us, Z_MIN_DELAY);
|
|
|
+ }
|
|
|
+ /// slow down
|
|
|
+ if(!_PINDA){
|
|
|
+ uint16_t steps_to_go = MAX(0, z - min_z);
|
|
|
+ while (!_PINDA && z > min_z){
|
|
|
+ go_and_stop(Z_AXIS_MASK, Z_ACCEL, current_delay_us, steps_to_go);
|
|
|
+ --z;
|
|
|
+ }
|
|
|
+ z_trig = z;
|
|
|
+ }
|
|
|
+ /// slow down to stop
|
|
|
+ while (z > min_z && current_delay_us < MAX_DELAY){
|
|
|
+ accelerate(Z_AXIS_MASK, -Z_ACCEL, current_delay_us, Z_MIN_DELAY);
|
|
|
+ --z;
|
|
|
}
|
|
|
- z_trig = z;
|
|
|
- z -= slow_down_z(delay_us);
|
|
|
|
|
|
count_position[2] = z;
|
|
|
if (d == 0){
|
|
@@ -735,7 +786,7 @@ bool xyzcal_scan_and_process(void){
|
|
|
uint8_t *matrix32 = (uint8_t *)block_buffer;
|
|
|
uint16_t *pattern = (uint16_t *)(matrix32 + 32 * 32);
|
|
|
|
|
|
- xyzcal_scan_pixels_32x32_Zhop(x, y, z - 72, 2400, 300, matrix32);
|
|
|
+ xyzcal_scan_pixels_32x32_Zhop(x, y, z - 72, 2400, 600, matrix32);
|
|
|
print_image(matrix32);
|
|
|
|
|
|
for (uint8_t i = 0; i < 12; i++){
|