klipper-dgus/scripts/motan/analyzers.py

284 lines
11 KiB
Python
Raw Normal View History

# Log data analyzing functions
#
# Copyright (C) 2021 Kevin O'Connor <kevin@koconnor.net>
#
# This file may be distributed under the terms of the GNU GPLv3 license.
import math, collections
import readlog
######################################################################
# Analysis code
######################################################################
# Analyzer handlers: {name: class, ...}
AHandlers = {}
# Calculate a derivative (position to velocity, or velocity to accel)
class GenDerivative:
ParametersMin = ParametersMax = 1
DataSets = [
('derivative(<dataset>)', 'Derivative of the given dataset'),
]
def __init__(self, amanager, name_parts):
self.amanager = amanager
self.source = name_parts[1]
amanager.setup_dataset(self.source)
def get_label(self):
label = self.amanager.get_label(self.source)
lname = label['label']
units = label['units']
if '(mm)' in units:
rep = [('Position', 'Velocity'), ('(mm)', '(mm/s)')]
elif '(mm/s)' in units:
rep = [('Velocity', 'Acceleration'), ('(mm/s)', '(mm/s^2)')]
else:
return {'label': 'Derivative', 'units': 'Unknown'}
for old, new in rep:
lname = lname.replace(old, new).replace(old.lower(), new.lower())
units = units.replace(old, new).replace(old.lower(), new.lower())
return {'label': lname, 'units': units}
def generate_data(self):
inv_seg_time = 1. / self.amanager.get_segment_time()
data = self.amanager.get_datasets()[self.source]
deriv = [(data[i+1] - data[i]) * inv_seg_time
for i in range(len(data)-1)]
return [deriv[0]] + deriv
AHandlers["derivative"] = GenDerivative
# Calculate an integral (accel to velocity, or velocity to position)
class GenIntegral:
ParametersMin = 1
ParametersMax = 3
DataSets = [
('integral(<dataset>)', 'Integral of the given dataset'),
('integral(<dataset1>,<dataset2>)',
'Integral with dataset2 as reference'),
('integral(<dataset1>,<dataset2>,<half_life>)',
'Integral with weighted half-life time'),
]
def __init__(self, amanager, name_parts):
self.amanager = amanager
self.source = name_parts[1]
amanager.setup_dataset(self.source)
self.ref = None
self.half_life = 0.015
if len(name_parts) >= 3:
self.ref = name_parts[2]
amanager.setup_dataset(self.ref)
if len(name_parts) == 4:
self.half_life = float(name_parts[3])
def get_label(self):
label = self.amanager.get_label(self.source)
lname = label['label']
units = label['units']
if '(mm/s)' in units:
rep = [('Velocity', 'Position'), ('(mm/s)', '(mm)')]
elif '(mm/s^2)' in units:
rep = [('Acceleration', 'Velocity'), ('(mm/s^2)', '(mm/s)')]
else:
return {'label': 'Integral', 'units': 'Unknown'}
for old, new in rep:
lname = lname.replace(old, new).replace(old.lower(), new.lower())
units = units.replace(old, new).replace(old.lower(), new.lower())
return {'label': lname, 'units': units}
def generate_data(self):
seg_time = self.amanager.get_segment_time()
src = self.amanager.get_datasets()[self.source]
offset = sum(src) / len(src)
total = 0.
ref = None
if self.ref is not None:
ref = self.amanager.get_datasets()[self.ref]
offset -= (ref[-1] - ref[0]) / (len(src) * seg_time)
total = ref[0]
src_weight = 1.
if self.half_life:
src_weight = math.exp(math.log(.5) * seg_time / self.half_life)
ref_weight = 1. - src_weight
data = [0.] * len(src)
for i, v in enumerate(src):
total += (v - offset) * seg_time
if ref is not None:
total = src_weight * total + ref_weight * ref[i]
data[i] = total
return data
AHandlers["integral"] = GenIntegral
# Calculate a kinematic stepper position from the toolhead requested position
class GenKinematicPosition:
ParametersMin = ParametersMax = 1
DataSets = [
('kin(<stepper>)', 'Stepper position derived from toolhead kinematics'),
]
def __init__(self, amanager, name_parts):
self.amanager = amanager
stepper = name_parts[1]
status = self.amanager.get_initial_status()
kin = status['configfile']['settings']['printer']['kinematics']
if kin not in ['cartesian', 'corexy']:
raise amanager.error("Unsupported kinematics '%s'" % (kin,))
if stepper not in ['stepper_x', 'stepper_y', 'stepper_z']:
raise amanager.error("Unknown stepper '%s'" % (stepper,))
if kin == 'corexy' and stepper in ['stepper_x', 'stepper_y']:
self.source1 = 'trapq(toolhead,x)'
self.source2 = 'trapq(toolhead,y)'
if stepper == 'stepper_x':
self.generate_data = self.generate_data_corexy_plus
else:
self.generate_data = self.generate_data_corexy_minus
amanager.setup_dataset(self.source1)
amanager.setup_dataset(self.source2)
else:
self.source1 = 'trapq(toolhead,%s)' % (stepper[-1:],)
self.source2 = None
self.generate_data = self.generate_data_passthrough
amanager.setup_dataset(self.source1)
def get_label(self):
return {'label': 'Position', 'units': 'Position\n(mm)'}
def generate_data_corexy_plus(self):
datasets = self.amanager.get_datasets()
data1 = datasets[self.source1]
data2 = datasets[self.source2]
return [d1 + d2 for d1, d2 in zip(data1, data2)]
def generate_data_corexy_minus(self):
datasets = self.amanager.get_datasets()
data1 = datasets[self.source1]
data2 = datasets[self.source2]
return [d1 - d2 for d1, d2 in zip(data1, data2)]
def generate_data_passthrough(self):
return self.amanager.get_datasets()[self.source1]
AHandlers["kin"] = GenKinematicPosition
# Calculate a toolhead x/y position from corexy stepper positions
class GenCorexyPosition:
ParametersMin = ParametersMax = 3
DataSets = [
('corexy(x,<stepper>,<stepper>)', 'Toolhead x position from steppers'),
('corexy(y,<stepper>,<stepper>)', 'Toolhead y position from steppers'),
]
def __init__(self, amanager, name_parts):
self.amanager = amanager
self.is_plus = name_parts[1] == 'x'
self.source1, self.source2 = name_parts[2:]
amanager.setup_dataset(self.source1)
amanager.setup_dataset(self.source2)
def get_label(self):
axis = 'x'
if not self.is_plus:
axis = 'y'
return {'label': 'Derived %s position' % (axis,),
'units': 'Position\n(mm)'}
def generate_data(self):
datasets = self.amanager.get_datasets()
data1 = datasets[self.source1]
data2 = datasets[self.source2]
if self.is_plus:
return [.5 * (d1 + d2) for d1, d2 in zip(data1, data2)]
return [.5 * (d1 - d2) for d1, d2 in zip(data1, data2)]
AHandlers["corexy"] = GenCorexyPosition
# Calculate a position deviation
class GenDeviation:
ParametersMin = ParametersMax = 2
DataSets = [
('deviation(<dataset1>,<dataset2>)', 'Difference between datasets'),
]
def __init__(self, amanager, name_parts):
self.amanager = amanager
self.source1, self.source2 = name_parts[1:]
amanager.setup_dataset(self.source1)
amanager.setup_dataset(self.source2)
def get_label(self):
label1 = self.amanager.get_label(self.source1)
label2 = self.amanager.get_label(self.source2)
if label1['units'] != label2['units']:
return {'label': 'Deviation', 'units': 'Unknown'}
parts = label1['units'].split('\n')
units = '\n'.join([parts[0]] + ['Deviation'] + parts[1:])
return {'label': label1['label'] + ' deviation', 'units': units}
def generate_data(self):
datasets = self.amanager.get_datasets()
data1 = datasets[self.source1]
data2 = datasets[self.source2]
return [d1 - d2 for d1, d2 in zip(data1, data2)]
AHandlers["deviation"] = GenDeviation
######################################################################
# Analyzer management and data generation
######################################################################
# Return a description of available analyzers
def list_datasets():
datasets = []
for ah in sorted(AHandlers.keys()):
datasets += AHandlers[ah].DataSets
return datasets
# Manage raw and generated data samples
class AnalyzerManager:
error = None
def __init__(self, lmanager, segment_time):
self.lmanager = lmanager
self.error = lmanager.error
self.segment_time = segment_time
self.raw_datasets = collections.OrderedDict()
self.gen_datasets = collections.OrderedDict()
self.datasets = {}
self.dataset_times = []
self.duration = 5.
def set_duration(self, duration):
self.duration = duration
def get_segment_time(self):
return self.segment_time
def get_datasets(self):
return self.datasets
def get_dataset_times(self):
return self.dataset_times
def get_initial_status(self):
return self.lmanager.get_initial_status()
def setup_dataset(self, name):
name = name.strip()
if name in self.raw_datasets:
return self.raw_datasets[name]
if name in self.gen_datasets:
return self.gen_datasets[name]
name_parts = readlog.name_split(name)
if name_parts[0] in self.lmanager.available_dataset_types():
hdl = self.lmanager.setup_dataset(name)
self.raw_datasets[name] = hdl
else:
cls = AHandlers.get(name_parts[0])
if cls is None:
raise self.error("Unknown dataset '%s'" % (name,))
num_param = len(name_parts) - 1
if num_param < cls.ParametersMin or num_param > cls.ParametersMax:
raise self.error("Invalid parameters to dataset '%s'" % (name,))
hdl = cls(self, name_parts)
self.gen_datasets[name] = hdl
self.datasets[name] = []
return hdl
def get_label(self, dataset):
hdl = self.raw_datasets.get(dataset)
if hdl is None:
hdl = self.gen_datasets.get(dataset)
if hdl is None:
raise self.error("Unknown dataset '%s'" % (dataset,))
return hdl.get_label()
def generate_datasets(self):
# Generate raw data
list_hdls = [(self.datasets[name], hdl)
for name, hdl in self.raw_datasets.items()]
initial_start_time = self.lmanager.get_initial_start_time()
start_time = t = self.lmanager.get_start_time()
end_time = start_time + self.duration
while t < end_time:
t += self.segment_time
self.dataset_times.append(t - initial_start_time)
for dl, hdl in list_hdls:
dl.append(hdl.pull_data(t))
# Generate analyzer data
for name, hdl in self.gen_datasets.items():
self.datasets[name] = hdl.generate_data()