mirror of https://github.com/Desuuuu/klipper.git
247 lines
11 KiB
Python
247 lines
11 KiB
Python
# Code for coordinating events on the printer toolhead
|
|
#
|
|
# Copyright (C) 2016 Kevin O'Connor <kevin@koconnor.net>
|
|
#
|
|
# This file may be distributed under the terms of the GNU GPLv3 license.
|
|
import math, logging, time
|
|
import cartesian
|
|
|
|
# Common suffixes: _d is distance (in mm), _v is velocity (in
|
|
# mm/second), _t is time (in seconds), _r is ratio (scalar between
|
|
# 0.0 and 1.0)
|
|
|
|
# Class to track each move request
|
|
class Move:
|
|
def __init__(self, toolhead, pos, move_d, axes_d, speed, accel):
|
|
self.toolhead = toolhead
|
|
self.pos = tuple(pos)
|
|
self.move_d = move_d
|
|
self.axes_d = axes_d
|
|
self.accel = accel
|
|
self.junction_max = speed**2
|
|
self.junction_delta = 2.0 * move_d * accel
|
|
self.junction_start_max = 0.
|
|
def calc_junction(self, prev_move):
|
|
# Find max start junction velocity using approximated
|
|
# centripetal velocity as described at:
|
|
# https://onehossshay.wordpress.com/2011/09/24/improving_grbl_cornering_algorithm/
|
|
junction_cos_theta = -((self.axes_d[0] * prev_move.axes_d[0]
|
|
+ self.axes_d[1] * prev_move.axes_d[1])
|
|
/ (self.move_d * prev_move.move_d))
|
|
if junction_cos_theta > 0.999999:
|
|
return
|
|
junction_cos_theta = max(junction_cos_theta, -0.999999)
|
|
sin_theta_d2 = math.sqrt(0.5*(1.0-junction_cos_theta));
|
|
R = self.toolhead.junction_deviation * sin_theta_d2 / (1. - sin_theta_d2)
|
|
self.junction_start_max = min(
|
|
R * self.accel, self.junction_max, prev_move.junction_max)
|
|
def process(self, junction_start, junction_end):
|
|
# Determine accel, cruise, and decel portions of the move
|
|
junction_cruise = self.junction_max
|
|
inv_junction_delta = 1. / self.junction_delta
|
|
accel_r = (junction_cruise-junction_start) * inv_junction_delta
|
|
decel_r = (junction_cruise-junction_end) * inv_junction_delta
|
|
cruise_r = 1. - accel_r - decel_r
|
|
if cruise_r < 0.:
|
|
accel_r += 0.5 * cruise_r
|
|
decel_r = 1.0 - accel_r
|
|
cruise_r = 0.
|
|
junction_cruise = junction_start + accel_r*self.junction_delta
|
|
self.accel_r, self.cruise_r, self.decel_r = accel_r, cruise_r, decel_r
|
|
# Determine move velocities
|
|
start_v = math.sqrt(junction_start)
|
|
cruise_v = math.sqrt(junction_cruise)
|
|
end_v = math.sqrt(junction_end)
|
|
self.start_v, self.cruise_v, self.end_v = start_v, cruise_v, end_v
|
|
# Determine time spent in each portion of move
|
|
accel_t = 2.0 * self.move_d * accel_r / (start_v + cruise_v)
|
|
cruise_t = self.move_d * cruise_r / cruise_v
|
|
decel_t = 2.0 * self.move_d * decel_r / (end_v + cruise_v)
|
|
self.accel_t, self.cruise_t, self.decel_t = accel_t, cruise_t, decel_t
|
|
# Generate step times for the move
|
|
next_move_time = self.toolhead.get_next_move_time()
|
|
self.toolhead.kin.move(next_move_time, self)
|
|
if self.axes_d[3]:
|
|
self.toolhead.extruder.move(next_move_time, self)
|
|
self.toolhead.update_move_time(accel_t + cruise_t + decel_t)
|
|
|
|
# Class to track a list of pending move requests and to facilitate
|
|
# "look-ahead" across moves to reduce acceleration between moves.
|
|
class MoveQueue:
|
|
def __init__(self):
|
|
self.queue = []
|
|
self.prev_junction_max = 0.
|
|
self.junction_flush = 0.
|
|
def flush(self, lazy=False):
|
|
next_junction_max = 0.
|
|
can_flush = not lazy
|
|
flush_count = len(self.queue)
|
|
junction_end = [None] * flush_count
|
|
for i in range(len(self.queue)-1, -1, -1):
|
|
move = self.queue[i]
|
|
junction_end[i] = next_junction_max
|
|
if not can_flush:
|
|
flush_count -= 1
|
|
next_junction_max = next_junction_max + move.junction_delta
|
|
if next_junction_max >= move.junction_start_max:
|
|
next_junction_max = move.junction_start_max
|
|
can_flush = True
|
|
prev_junction_max = self.prev_junction_max
|
|
for i in range(flush_count):
|
|
move = self.queue[i]
|
|
next_junction_max = min(prev_junction_max + move.junction_delta
|
|
, junction_end[i])
|
|
move.process(prev_junction_max, next_junction_max)
|
|
prev_junction_max = next_junction_max
|
|
del self.queue[:flush_count]
|
|
self.prev_junction_max = prev_junction_max
|
|
self.junction_flush = 0.
|
|
if self.queue:
|
|
self.junction_flush = self.queue[-1].junction_max
|
|
def add_move(self, move):
|
|
self.queue.append(move)
|
|
if len(self.queue) == 1:
|
|
self.junction_flush = move.junction_max
|
|
return
|
|
move.calc_junction(self.queue[-2])
|
|
self.junction_flush -= move.junction_delta
|
|
if self.junction_flush <= 0.:
|
|
self.flush(lazy=True)
|
|
|
|
STALL_TIME = 0.100
|
|
|
|
# Main code to track events (and their timing) on the printer toolhead
|
|
class ToolHead:
|
|
def __init__(self, printer, config):
|
|
self.printer = printer
|
|
self.reactor = printer.reactor
|
|
self.extruder = printer.objects.get('extruder')
|
|
self.kin = cartesian.CartKinematics(printer, config)
|
|
self.max_xy_speed, self.max_xy_accel = self.kin.get_max_xy_speed()
|
|
self.junction_deviation = config.getfloat('junction_deviation', 0.02)
|
|
self.move_queue = MoveQueue()
|
|
self.commanded_pos = [0., 0., 0., 0.]
|
|
# Print time tracking
|
|
self.buffer_time_high = config.getfloat('buffer_time_high', 5.000)
|
|
self.buffer_time_low = config.getfloat('buffer_time_low', 0.150)
|
|
self.move_flush_time = config.getfloat('move_flush_time', 0.050)
|
|
self.motor_off_delay = config.getfloat('motor_off_time', 60.000)
|
|
self.print_time = 0.
|
|
self.print_time_stall = 0
|
|
self.motor_off_time = self.reactor.NEVER
|
|
self.flush_timer = self.reactor.register_timer(self.flush_handler)
|
|
def build_config(self):
|
|
self.kin.build_config()
|
|
# Print time tracking
|
|
def update_move_time(self, movetime):
|
|
self.print_time += movetime
|
|
flush_to_time = self.print_time - self.move_flush_time
|
|
self.printer.mcu.flush_moves(flush_to_time)
|
|
def get_next_move_time(self):
|
|
if not self.print_time:
|
|
self.print_time = self.buffer_time_low + STALL_TIME
|
|
curtime = time.time()
|
|
self.printer.mcu.set_print_start_time(curtime)
|
|
self.reactor.update_timer(self.flush_timer, self.reactor.NOW)
|
|
return self.print_time
|
|
def get_last_move_time(self):
|
|
self.move_queue.flush()
|
|
return self.get_next_move_time()
|
|
def reset_motor_off_time(self, eventtime):
|
|
self.motor_off_time = eventtime + self.motor_off_delay
|
|
def reset_print_time(self):
|
|
self.move_queue.flush()
|
|
self.printer.mcu.flush_moves(self.print_time)
|
|
self.print_time = 0.
|
|
self.reset_motor_off_time(time.time())
|
|
self.reactor.update_timer(self.flush_timer, self.motor_off_time)
|
|
def check_busy(self, eventtime):
|
|
if not self.print_time:
|
|
# XXX - find better way to flush initial move_queue items
|
|
if self.move_queue.queue:
|
|
self.reactor.update_timer(self.flush_timer, eventtime + 0.100)
|
|
return False
|
|
buffer_time = self.printer.mcu.get_print_buffer_time(
|
|
eventtime, self.print_time)
|
|
return buffer_time > self.buffer_time_high
|
|
def flush_handler(self, eventtime):
|
|
if not self.print_time:
|
|
self.move_queue.flush()
|
|
if not self.print_time:
|
|
if eventtime >= self.motor_off_time:
|
|
self.motor_off()
|
|
self.reset_print_time()
|
|
self.motor_off_time = self.reactor.NEVER
|
|
return self.motor_off_time
|
|
print_time = self.print_time
|
|
buffer_time = self.printer.mcu.get_print_buffer_time(
|
|
eventtime, print_time)
|
|
if buffer_time > self.buffer_time_low:
|
|
return eventtime + buffer_time - self.buffer_time_low
|
|
self.move_queue.flush()
|
|
if print_time != self.print_time:
|
|
self.print_time_stall += 1
|
|
self.dwell(self.buffer_time_low + STALL_TIME)
|
|
return self.reactor.NOW
|
|
self.reset_print_time()
|
|
return self.motor_off_time
|
|
def stats(self, eventtime):
|
|
buffer_time = 0.
|
|
if self.print_time:
|
|
buffer_time = self.printer.mcu.get_print_buffer_time(
|
|
eventtime, self.print_time)
|
|
return "print_time=%.3f buffer_time=%.3f print_time_stall=%d" % (
|
|
self.print_time, buffer_time, self.print_time_stall)
|
|
# Movement commands
|
|
def get_position(self):
|
|
return list(self.commanded_pos)
|
|
def set_position(self, newpos):
|
|
self.move_queue.flush()
|
|
self.commanded_pos[:] = newpos
|
|
self.kin.set_position(newpos)
|
|
def _move_with_z(self, newpos, axes_d, speed):
|
|
self.move_queue.flush()
|
|
move_d = math.sqrt(sum([d*d for d in axes_d[:3]]))
|
|
# Limit velocity and accel to max for each stepper
|
|
kin_speed, kin_accel = self.kin.get_max_speed(axes_d, move_d)
|
|
speed = min(speed, self.max_xy_speed, kin_speed)
|
|
accel = min(self.max_xy_accel, kin_accel)
|
|
# Generate and execute move
|
|
move = Move(self, newpos, move_d, axes_d, speed, accel)
|
|
move.process(0., 0.)
|
|
def _move_only_e(self, newpos, axes_d, speed):
|
|
self.move_queue.flush()
|
|
kin_speed, kin_accel = self.extruder.get_max_speed()
|
|
speed = min(speed, self.max_xy_speed, kin_speed)
|
|
accel = min(self.max_xy_accel, kin_accel)
|
|
move = Move(self, newpos, abs(axes_d[3]), axes_d, speed, accel)
|
|
move.process(0., 0.)
|
|
def move(self, newpos, speed, sloppy=False):
|
|
axes_d = [newpos[i] - self.commanded_pos[i]
|
|
for i in (0, 1, 2, 3)]
|
|
self.commanded_pos[:] = newpos
|
|
if axes_d[2]:
|
|
self._move_with_z(newpos, axes_d, speed)
|
|
return
|
|
move_d = math.sqrt(axes_d[0]**2 + axes_d[1]**2)
|
|
if not move_d:
|
|
if axes_d[3]:
|
|
self._move_only_e(newpos, axes_d, speed)
|
|
return
|
|
# Common xy move - create move and queue it
|
|
speed = min(speed, self.max_xy_speed)
|
|
move = Move(self, newpos, move_d, axes_d, speed, self.max_xy_accel)
|
|
self.move_queue.add_move(move)
|
|
def home(self, axis):
|
|
return self.kin.home(self, axis)
|
|
def dwell(self, delay):
|
|
self.get_last_move_time()
|
|
self.update_move_time(delay)
|
|
def motor_off(self):
|
|
self.dwell(STALL_TIME)
|
|
last_move_time = self.get_last_move_time()
|
|
self.kin.motor_off(last_move_time)
|
|
self.extruder.motor_off(last_move_time)
|
|
self.dwell(STALL_TIME)
|
|
logging.debug('; Max time of %f' % (last_move_time,))
|