AI generated capacitor list ... nope

[Klondike Mike]

I spent the afternoon in preparation for building a new kit for a radio. I wanted to find out if AI would be able to remove one of the hurdles in my usual tasks that must be performed when building a new kit. I initially asked Google's AI to create a list for me. I used a service manual it found online. The end result was utter failure. Overblown capacitor count by 40%, values not even listed, wrong values listed, board # designations mostly omitted. This was after multiple attempts and disputing its findings as I went through the service manual BOM. All of the data generated is just slop and will be discarded.

I wasted a lot of time with this AI but with the results provided, I felt re-assured that a human is still required to build an accurate list that can be turned into a physical kit. A human is still needed to pick/order/package/ship etc., so AI is not going to be taking over my job anytime soon.

 

[TM007]

Let me say first that you provide a great service and although I have never had one of your kits myself I think the many folk that have who testament to them says enough.

As it was not viable due to the ocean between us and ridiculous postage costs upon which you have no control, I was not able to purchase from you earlier this year. But you were gracious in taking your time to reply to my enquiry at the time.

This left me with sourcing capacitors for a number of radio's which I have posted about on this forum. Rather than repeat the detail here that thread is and the post in particular is AI the good, bad & ugly CAPS .
Suffice to say there is another thread regarding AI also which documents some frustrations and some solutions or ways of thinking when approaching AI.

First thing is to determine if you were using Gemini (Google AI) as part of a paid subscription specifically enhancing the AI experience or just the out the box AI experience?

Next would be understanding the ground rules you set Gemini to T it up for the task you set it, is that something you could share?

Further to that would understanding the prompts you used along your journey?

Suffice to say when approached and implemented well AI can be very useful and was very useful when I wanted to collate my BOM and upload to supplier.
I think it comes down to necessity (being the Mother of Invention, Afterall). If you want to make your commercial aspect a little leaner, maybe make a little more and even pass on some reductions to buyers then it might all be around the upfront work to get the AI process slick enough to work to that end. Also has to be said that very little that's good or great is free, so paying for better AI to do more complex tasks to benefit your business may or may not make sense.

That's a key aspect is right there! the upfront work, repeatable, designed to be improved easily and potentially working across multiple suppliers. You're also going to have to consider the source material (the service manuals etc.) are they good readable clean OCR or just pdf pictures of jpg's and the like! the source material needs to be AI readable and thats not free (for the good stuff that is) and if its poor then your going to get poor output. Fixing the documents with paid for AI services that can actually give you better readable data probably comes before anything.

As was said in that other thread G.I.G.O.

Its a chore to setup but if you do it correctly and put in the initial time its occasionally worth it.
What I would say lastly is you could be able to drop one of the 4 human aspects you mentioned "pick/order/package/ship" order! if the AI works for you. But the other 3 probably aren't worth the effort for the quantities and cost in your model (hazarding a guess there, no offence intended).

 

[nomadradio]

I wasted a lot of time with this AI

Well look on the bright side. You trained it for free.

73

 

[brandon7861]

There is a list of AI sites that specialize in electronics. Flux, ProtoFlow, Circuit Mind, DeepPCB, Celus, Quilter, Cadence Allegro X, etc. Maybe one of those would be better at recognizing electrolytic cap symbols.

 

[Klondike Mike]

Let me say first that you provide a great service and although I have never had one of your kits myself I think the many folk that have who testament to them says enough... First thing is to determine if you were using Gemini (Google AI) as part of a paid subscription specifically enhancing the AI experience or just the out the box AI experience?

Appreciate your detailed response. I just sent a few kits out to Ireland (from western Canada), delivered in 6 days with regular Canada Post shipping. I can't get that delivery time for some eastern Canada addresses. Any-who ...

I just used the AI that Google provides when doing a search. No paid subscription or any special account access, only what a regular joe would use. I'm sure we could increase the accuracy rate by smoothing out the rough edges on this exercise with additional training and cleaner data but what about the next one? I am not looking for an answer as the effort required for each unique one off situation would not be worth it.

Public link to Google AI conversation, click here

 

[Klondike Mike]

There is a list of AI sites that specialize in electronics. Flux, ProtoFlow, Circuit Mind, DeepPCB, Celus, Quilter, Cadence Allegro X, etc. Maybe one of those would be better at recognizing electrolytic cap symbols.

Yikes! How many frick'in AI databases are out there?

In my case, all Google had to do was OCR the manual (which is a pretty good copy) and read a couple pages of text (which it did rather quickly). A simple task that humans can easily do. OCR technology has been around for many years and last time I used it was pretty good.

 

[TM007]

Appreciate your detailed response. I just sent a few kits out to Ireland (from western Canada), delivered in 6 days with regular Canada Post shipping. I can't get that delivery time for some eastern Canada addresses. Any-who ...

I just used the AI that Google provides when doing a search. No paid subscription or any special account access, only what a regular joe would use. I'm sure we could increase the accuracy rate by smoothing out the rough edges on this exercise with additional training and cleaner data but what about the next one? I am not looking for an answer as the effort required for each unique one off situation would not be worth it.

Public link to Google AI conversation, click here

Yeah wasn't the speed of delivery if you remember it was the cost of the postage (not your fault), would have been ok with a couple of weeks at the time. Its all good though.

So, upfront its hard to unsee what you've seen so to speak, a bit like that Tom Hanks "Sully" movie re the simulator pilots taking the human element out of the situation. I've seen what you wrote but if as a person interested in Radio and recapping, its reasonable that I would seek out the Service Manual and download it locally, it also reasonable that I would determine the Parts list were on page 3>9 inclusive and ignore the rest of the pages initially.

So I write up a quick Prompt takes around 5 mins. Its not great as prompts go but the gist is there. looks like this:
Acting as a RF enginner looking to recap electrolytic capacitors only on a Alinco DR-570T/E. I'd like an accurate spreadsheet that provides the following information in respect of replacment electrolytic capacitors of the same specification or better in regards to voltage. Also pay attention to dimension of capacitor where possible, again the same dimension or smaller as found in the attached parts list.

Use additinal headings like Board Designation e.g. "VHF Main UNIT" the count required for each capacitor and all other relevant values required for providing a B.O.M. to the suppliers below.

I require you to verify your findings before creating any spreadsheet. Please use the following upload from the Alinco DR-570T/E Service manual, this is a parts list only. Verify you can OCR this accurately and highlight any aspects that may need refinement to be able to provide the initial spreadsheet.

Use suppliers like Farnell, specifically https://cpc.farnell.com/c/electroni...tors/leaded-aluminium-electrolytic-capacitors DigiKey https://www.digikey.com/en/products/filter/aluminum-electrolytic-capacitors/ and Mouser https://www.mouser.com/en/c/passive-components/capacitors/aluminum-electrolytic-capacitors/ provide a sheet for each of these suppliers within the spreadsheet. If possible also use any critiria specified on those sites as regards the uploading on a B.O.M. making each sheet compatible with that individuals sites B.O.M. upload facility.‌

Then i use an AI tool to improve that Prompt Prompt Cowboy can be useful its free.
Now it looks like this:
You are an RF engineer sourcing a recap kit of electrolytic capacitors only for an Alinco DR-570T/E transceiver, working from the attached parts list extracted from the Alinco DR-570T/E service manual.

Before producing any spreadsheet, OCR and verify the attached parts list document. Report back explicitly:
- Confirmation of which entries you could read with high confidence versus any that are ambiguous, low-resolution, or contradictory (e.g. unclear voltage ratings, unclear capacitance values, unclear board designations, or unclear physical dimensions).
- Any parts list entries where the component type is ambiguous (i.e. it's unclear whether it's electrolytic vs. another capacitor type) — since only electrolytic capacitors are in scope.
- Any gaps where board designation, quantity, or dimension data is missing or unclear.

Flag all of this before generating the spreadsheet, so I can confirm or correct anything uncertain.

Once verified, produce a spreadsheet covering only the electrolytic capacitors from that parts list, structured as a Bill of Materials for submission to suppliers. For each capacitor, include:
- Board Designation (e.g. "VHF Main Unit") as given in the parts list
- Original reference/part number as listed
- Original capacitance and voltage rating
- Replacement capacitance and voltage rating — same specification or better voltage rating (equal or higher voltage rated), matching original capacitance
- Replacement physical dimensions — same size or smaller than the original where a suitable replacement exists, and note explicitly if no smaller/equal option was found
- Quantity required for that specific capacitor
- Any other value relevant to sourcing and B.O.M. submission (e.g. temperature rating, lead pitch/spacing, manufacturer part number)

Source verified, currently available replacement parts from Farnell/CPC (https://cpc.farnell.com/c/electroni...tors/leaded-aluminium-electrolytic-capacitors), DigiKey (https://www.digikey.com/en/products/filter/aluminum-electrolytic-capacitors/), and Mouser (https://www.mouser.com/en/c/passive-components/capacitors/aluminum-electrolytic-capacitors/). Provide the B.O.M. as a separate sheet for each supplier, and within each sheet, format columns and headers to match that supplier's specific B.O.M. upload template requirements as closely as possible — check each supplier's site for their required upload format (e.g. required column names, order, and any mandatory fields) and structure the sheet accordingly. Note explicitly any supplier-specific formatting requirement you couldn't confirm or apply.

For every replacement part, cite the exact manufacturer part number and supplier listing used, and flag any capacitor where you could not find a replacement meeting the voltage and dimension criteria, explaining why.


BTW I got 39 Caps but that included ones for option boards... I didn't count them myself. Each attached workbook contains 3 tabs all contain relevant information to the list achieved.
The Service Manual from the 2 sources provided were not great from an OCT aspect, they were ok but I had to run it through PDF24 OCR and Best quality and even then there was ambiguity that was outlined though.

I can't share my AI "Chat" as you helpfully did due to some restrictions. However I can share the intial output to the Initial prompt.. Apologies for the long post test here.

Next post....as word count too high... apologies.

 

[TM007]

RESPONSE:
I completed an OCR pass and checked the extracted text against the scanned tables in Alinco. I have not generated the spreadsheet or selected replacement parts yet, because the source list does not provide physical dimensions, lead spacing, or an explicit quantity column, and a few entries need confirmation.

1. Scope and component-type interpretation​

The parts list uses“Chemical C” for the capacitors that appear to be conventional aluminium electrolytics. I propose treating every “Chemical C” entry as in scope. The document does not literally say “aluminium electrolytic”, so this is a type interpretation rather than an explicit statement.

I have excluded:

• “Chip Tantal” entries, because these are tantalum capacitors.
• “Chip C”, ceramic, mica, film and trimmer capacitors.
• Entries whose capacitance is in pF and whose description identifies another dielectric.

There are no entries where the printed description oscillates between “Chemical C” and “Chip Tantal”. The main scope ambiguity is therefore whether you want all “Chemical C” parts treated as aluminium electrolytics.

2. High-confidence electrolytic entries​

The following transcription is high confidence from the scan. Each reference appears once in the list, but the document does not provide a formal quantity field.

VHF Main Unit​

• C5E, part code CE0200, Chemical C, UW 50 V 2.2 µF, DR-570E only
• C8E, part code CE0189, Chemical C, TWSS 100 µF/16 V, DR-570E only
• C40, part code CE0188, Chemical C, TWSS 47 µF/16 V
• C48, part code CE0189, Chemical C, TWSS 100 µF/16 V
• C49, part code CE0188, Chemical C, TWSS 47 µF/16 V
• C50, part code CE0188, Chemical C, TWSS 47 µF/16 V
• C52, part code CE0188, Chemical C, TWSS 47 µF/16 V
• C58, part code CE0188, Chemical C, TWSS 47 µF/16 V
• C59, part code CE0181, Chemical C, TWSS 470 µF/10 V
• C84, part code CE0188, Chemical C, TWSS 47 µF/16 V
• C90, part code CE0178, Chemical C, TWSS 220 µF/10 V
• C93, part code CE0078, Chemical C, MS9 100 µF/25 V
• C100, part code CE0044, Chemical C, MSS 10 µF/16 V

UHF Main Unit​

• C34, part code CE0188, Chemical C, TWSS 47 µF/16 V
• C72, part code CE0200, Chemical C, UW 50 V 2.2 µF
• C79, part code CE0188, Chemical C, TWSS 47 µF/16 V
• C83, part code CE0178, Chemical C, TWSS 220 µF/10 V
• C103, part code CE0188, Chemical C, TWSS 47 µF/16 V
• C119, part code CE0189, Chemical C, TWSS 100 µF/16 V
• C121, part code CE0044, Chemical C, MS5 10 µF/16 V
• C128, part code CE0044, Chemical C, MS5 10 µF/16 V
• C137, part code CE0044, Chemical C, MS5 10 µF/16 V
• C140, part code CE0188, Chemical C, TWSS 47 µF/16 V

One additional UHF Main entry, C82, is withheld from this high-confidence group because of the capacitance-unit contradiction described below.

CPU Unit​

• C3, part code CE0037, Chemical C, MS5 100 µF/6 V
• C4, part code CE0037, Chemical C, MS5 100 µF/6 V
• C11, part code CE0037, Chemical C, MSS 100 µF/6 V

The board association is readable from the column layout, although the OCR text stream itself interleaves adjacent columns.

Control Unit​

• C13, part code CE0097, Chemical C, MS5 68 µF/10 V
• C49, part code CE0097, Chemical C, MS5 68 µF/10 V
• C55, part code CE0198, Chemical C, MS5 100 µF/10 V
• C56, part code CE0198, Chemical C, MS5 100 µF/10 V

Power Source Unit​

• C78, part code CE0193, Chemical C, TWSS 1000 µF/16 V

UHF Final Unit​

The following entries are readable as Chemical C, MS5, 10 µF/16 V:

• C7, part code CE0044
• C13, part code CE0044
• C14, part code CE0044
• C15, part code CE0044
• C45, part code CE0044
• C48, part code CE0044
• C52, part code CE0044
• C54, part code CE0044

CTCSS EJ-5U Unit, T model only​

• C5, part code CE0116, Chemical C, MS7 6.3 V/47 µF

The printed order is unusual, but readable as 47 µF, 6.3 V.

3. Entries requiring confirmation​

UHF Main Unit C82​

The scan/OCR gives:

• Reference: C82
• Part code: CE0178
• Description: Chemical C, TWSS 220 pF/10 V

This is internally contradictory. The adjacent C83 uses the same CE0178 code and is printed as 220 µF/10 V. A chemical electrolytic marked 220 pF would be implausible, but I will not silently correct the source. My likely transcription is 220 µF/10 V, subject to your confirmation.

UHF Final Unit C53/C54 area​

The scan’s column alignment around C52-C54 is difficult, and the OCR produces one reading of“10 pF/16 V” among otherwise repeated CE0044 Chemical C entries. The visual table appears to associate C54 with CE0044 and 10 µF/16 V, but this should be confirmed before purchasing.

Series-name inconsistencies​

The scan includes visually similar series labels:

• MS5
• MS7
• MS9
• MSS
• TWSS
• UW

These are legible enough to transcribe, but the document gives no series datasheets or manufacturer identity. In particular, MS5 versus MSS could be important if these codes encode construction or dimensions.

4. Missing quantity information​

The source has columns for:

• Reference number
• Part code
• Part name and number

It has no quantity column. My proposed BOM treatment is:

• One unit for each distinct reference designation.
• Consolidated purchasing quantity equal to the number of references using the same replacement part.
• DR-570E-only parts kept separately flagged rather than included automatically in a DR-570T kit.
• The CTCSS EJ-5U capacitor kept as an optional T-model accessory-board item.

This quantity treatment is an engineering assumption and is not directly stated by the parts list.

5. Missing physical-dimension and lead-spacing information​

The attached parts list provides no explicit case diameter, body height, lead pitch, lead diameter or mounting style for any Chemical C entry. The series-like labels such as TWSS, MS5 and MS7 are not sufficient by themselves to establish dimensions without a corresponding legacy datasheet.

Consequently, I cannot yet truthfully certify that any supplier replacement is “same size or smaller” solely from this document. For the spreadsheet, the honest options would be:

1. Mark every original dimension as“Not stated in supplied parts list” and compare replacements only after you provide measured dimensions; or
2. Research legacy datasheets for CE0188, CE0189, CE0178, CE0044 and the other Alinco codes, while clearly identifying any dimension derived from an external datasheet rather than the supplied list.

Similarly, the original list does not state:

• Lead pitch or spacing
• Radial versus axial configuration
• Temperature rating
• Endurance or lifetime
• ESR or impedance class
• Ripple-current rating
• Tolerance
• Polarity orientation
• Maximum body height imposed by the chassis

6. Confirmation needed before supplier sourcing​

Please confirm these four points:

1. Treat every“Chemical C” entry as an aluminium electrolytic in scope.
2. Interpret UHF Main C82 CE0178 as 220 µF/10 V, not 220 pF/10 V.
3. Accept one capacitor per reference designation where the list has no quantity column.
4. Either provide measured original dimensions, or approve a BOM that marks original dimensions as unavailable and flags dimensional compliance as unverified until those measurements are supplied.

Once those points are resolved, I can research current Farnell/CPC, DigiKey and Mouser listings and build the supplier-specific workbook without embedding unsupported dimensional assumptions.

Next post....

 

[TM007]

Further to that it called out the following:

After re-reading the OCR output, I am not confident that 41 is a verified count. The problem is that pages 2-7 are three-column parts tables and the OCR stream interleaves columns. A number of the electrolytic entries I carried into the workbook were inferred from OCR alignment rather than independently verified visually from the scanned page.
At this stage I would classify the capacitor count as:

High-confidence verified references​

Board	Count
VHF Main Unit	13
UHF Main Unit	10 confirmed, plus 1 questionable (C82)
CPU Unit	3
Control Unit	4
Power Source Unit	1
CTCSS EJ-5U Unit	1
Subtotal	32 confirmed + 1 questionable

That gives me 32 definite electrolytic references, or 33 if C82 is accepted as an electrolytic capacitor.
The area I am least happy with is the UHF Final Unit, where the OCR stream becomes heavily merged and several capacitors were carried over because they appeared to be CE0044 "Chemical C" entries. From the OCR text alone I cannot honestly prove all eight UHF Final references (C7, C13, C14, C15, C45, C48, C52, C54) belong to that board and are all electrolytics.
If all eight UHF Final entries are removed, the total would drop from 41 to:

• 41 - 8 = 33

which aligns very closely with my independently verified count above.
So my current confidence assessment is:

Count	Confidence
41 capacitors	Low
33 capacitors	Medium
32 capacitors	High if C82 excluded

The correct engineering approach is not to continue building DigiKey or Mouser BOMs from the 41-count workbook.
I recommend I perform a second-pass visual audit of the PDF pages and rebuild the electrolytic list from the page images only, producing:

1. Confirmed electrolytics.
2. Ambiguous electrolytics.
3. Excluded entries.
4. Final verified quantity.

Only then should we freeze the Farnell BOM and use it as the basis for DigiKey and Mouser.
At the moment, I would not sign off the 41-capacitor figure as verified. My best estimate is that the true count is likely 32-33 capacitors, but I cannot yet state the exact number without a fresh visual pass through the scanned pages.

This is where I cleaned up the original PDF with PDF 24 (FREE). and uploaded a cleaner copy.

rebuilt the list directly from the seven cleaned page images, reading each table column separately and counting only entries explicitly labelled“Chemical C”. I count 39 electrolytic-capacitor entries in the complete parts list, with high confidence.

Verified count by board​

Board designation	Verified references	Count
VHF Main Unit	C5E, C8E, C40, C48, C49, C50, C52, C58, C59, C84, C90, C93, C100	13
UHF Main Unit	C34, C72, C79, C82, C83, C103, C119, C121, C128, C137, C140	11
CPU Unit	C3, C4, C11	3
Control Unit	C13, C49, C55, C56	4
Power Source Unit	C78	1
UHF Final Unit	C7, C13, C14, C15, C45, C48	6
CTCSS EJ-5U Unit	C5	1
Complete parts-list total		39

Arithmetic check​

13 + 11 + 3 + 4 + 1 + 6 + 1 = 39 capacitors.

What was wrong in the 41-row spreadsheet​

The previous spreadsheet incorrectly included two additional UHF Final Unit references:

• C52 is not electrolytic. The cleaned table identifies C52 as CU0001, Chip C, C2012CH1H 0.5 pF.
• C54 is not electrolytic. The table identifies C54 as a chip capacitor, not a CE0044 Chemical C component.

Removing those two incorrect rows changes the count from 41 to 39.

Important cleaned-list corrections​

The cleaned document also resolves several earlier OCR problems:

• UHF Main C82 is visibly CE0178, Chemical C, TWSS 220 µF/10 V.
• UHF Main C83 has the same part code and specification, although one OCR text layer still mistranscribes the unit as pF. The visual table supports 220 µF/10 V.
• VHF Main C93 is visually a Chemical C entry. The OCR text layer reports “100 pF/25 V”, but the capacitor type and printed unit appearance support 100 µF/25 V.
• UHF Main C121 is Chemical C, MS5, 10 µF/16 V, despite one OCR rendering of “10 pF”.
• Control Unit C55 is Chemical C, MS5, 100 µF/10 V, despite one OCR rendering of “100 pF”.
• UHF Final C54 must be excluded, rather than assumed to be 10 µF/16 V.

Model-specific ambiguity​

The count of 39 is the number of Chemical C entries across the complete DR-570T/E parts list, including model-specific and optional-board entries. It is not necessarily the installed count in one individual radio.
Two VHF Main Unit entries are explicitly E-only:

• C5E
• C8E

The CTCSS section is headed“CTCSS (EJ-5U) Unit (T Only)”, but the C5 Chemical C row appears to carry“DR-570E only” in the table. Those statements contradict each other, so I cannot confidently assign that capacitor to a T-only or E-only kit without another source. I can confidently say it is one Chemical C entry in the complete list.
Therefore:

• Complete master parts-list count: 39
• Excluding the optional/contradictory CTCSS entry: 38
• Excluding CTCSS and the two explicitly E-only VHF capacitors: 36 common/non-E-suffix entries
• A definitive installed count for a particular DR-570T or DR-570E requires resolving the CTCSS applicability contradiction and confirming whether that option board is fitted.