JLCPCB 2-side assembly rotation...

[rokan51]

Hi, I am using kikit for JLCPCB fabrication and 2 side assembly. The top side looks all good, however bottom side generates in POS file positions which some components wrongly rotated when uploaded to JLCPCB site: it looks to me that the rotation orientation is wrong - elements rotated by 0, 180 deg are good but 90 deg rotated are wrong (opposite than they should be, all my rotations are multiple of 90). Also correction field seems to have swapped X/Y axes when placed on the bottom side... any suggestions?

[yaqwsx]

It seems that JLC PCB has changed the interpretation of the POS files. We need to address this in KiKit.

[rokan51]

Right, it seems that in kicad flipping component from top to bottom side means rotating around local (component-coordinate) Y axis whereas for JLCPCB changing T->B (with the same coordinates) means rotating around local X axis. From pos.csv generated by your kikit fabrication I have further applied following correction formula, which can be expressed in pseudo-C:
Rotation = (Layer=="B") ? 180 - Rotation : Rotation;
So far I have tried it in LibreCalc using
=IF(Dx="B", Ex*(1-)+180, Ex)
and replacing original Rotation column with such an update, uploading to JLCPCB... everything seems fine!

OT PS: It might be useful to allow correction to be loaded not only from component fields but from file(s) /database(multiple-files) where correction would be specified according to package name (with library prefix, of course). In such a way board files would not need to be modified each time, users could share their correction files, a repository of such shared corrections can be created... (still "field correction" could be used to override the database for extra tuning) e.g. using some extra option like --correction-files "/some/place/*" or so.
Anyway great work, thanks a lot!

[rokan51]

Ok, here is a short "C-script" (I am running that using tcc) which corrects bottom layer. It reads from stdin, writes to stdout so it could be run like
cat pos,scv | jlcpcb-correct > pos-c.scv

#!/usr/bin/tcc -run

#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <ctype.h>

char buff[1024];
int line = 1;

void scan_comma(){
  int e = scanf("%[^,],", buff);
  if(e<0) exit(0); // EOF
  if(e!=1){
     fprintf(stderr, "Error on line %d\n.", line);
     exit(1);
  }
}
int scan_line(){
  int e = scanf("%[^\n]%*c", buff);
  if(e!=1){
    fprintf(stderr, "Error on line %d\n.", line);
    exit(2);
  }
  line++;
}

double get_value(){
  if(strlen(buff)<1){
    fprintf(stderr, "Bad value conversion on line %d\n.", line);
    exit(3);
  }
  return atof(buff);
}
char get_first(){
  const char * b = buff;
  while(isspace(*b)) b++;
  return *b;
}

int main(){
  double x, y, rot;
  char layer;
  scan_line();
  printf("%s\n", buff); // forwarding header to stdout
  while(1){
    scan_comma(); printf("%s,", buff); // designator
    scan_comma();  x = get_value();
    scan_comma();  y = get_value();
    scan_comma();  layer = get_first();
    scan_line(); rot = get_value();
    rot = (tolower(layer)=='b') ?  180 - rot : rot;
    printf("%g,%g,%c,%g\n", x, y, layer, rot); 
  }
  return 0;
}

[markh-de]

Thanks, @rokan51, for your analysis. The fix is really as simple as 180 − angle.

Actually, it's a bit more complicated, as the orientation compensation data needs to be taken into account. Unfortunately, this data is not available in the resulting CSV file, so we cannot really work around this with simple post-processing.

Do not use the following script. It cannot handle the compensation angle!

For anyone who prefers Python, this is a simple fixup script that you need to run exactly once on the position file generated by KiKit:

#!/usr/bin/env python

import csv
import sys
import os

""" Works around https://github.com/yaqwsx/KiKit/discussions/664 """

def recalculate_rotation(value): return (180 - float(value)) % 360

def fix_rotation(file_name):
    with open(file_name, mode='r', newline='', encoding='utf-8') as file:
        reader = csv.DictReader(file)
        rows = list(reader)
        fieldnames = reader.fieldnames

    if not fieldnames or "Layer" not in fieldnames or "Rotation" not in fieldnames:
        print("Error: CSV file must contain 'Layer' and 'Rotation' columns.")
        return

    for row in rows:
        if row["Layer"] == "B": row["Rotation"] = recalculate_rotation(row["Rotation"])

    with open(file_name, mode='w', newline='', encoding='utf-8') as file:
        writer = csv.DictWriter(file, fieldnames=fieldnames)
        writer.writeheader()
        writer.writerows(rows)

if __name__ == "__main__":
    script_name = os.path.basename(sys.argv[0])
    if len(sys.argv) < 2:
        print(f"Usage: {script_name} <pos_csv_file>")
    else:
        fix_rotation(sys.argv[1])

[yaqwsx]

Have you considered submitting a 2 line PR that would fix the issue?

I have the same observation, but I struggle to get a proper specification from JLC PCB how the components should be oriented for each side.

[markh-de]

It's a few more changes, since I discovered that we need to handle the orientation compensation angle, too. Otherwise (as in the scripts above - beware!) the correction only works for parts with compensation angle == 0.

Please have a look at #767.

The PR implementation was tested against the assembly preview at JLCPCB for a double-sided board with some identical parts (with same compensation data) on both sides and some weird component angles (like 13°, 167°). For me, it works for all tested cases.

Please test for yourselves.