Modbus Spindle Control

NineMile

@gloomyandy said in Modbus Spindle Control:

I don't think that is really the case. If you take a look at the modbus write funtion here: https://github.com/Duet3D/RepRapFirmware/blob/3.6-dev/src/Comms/AuxDevice.cpp#L190

Ah yep, good point on the timing concerns. So the user on Discord who has used the UART functions to implement the HY style "Modbus" comms might just be getting lucky with timing / silent durations that work for that particular VFD but with no guarantee of working on others.

@gloomyandy said in Modbus Spindle Control:

So I think there might still be a case for having modbus specific read/write operations but I'd be tempted to try and make them more generic ones that simply read/write a bunch of bytes and apply the timing and crc to them, but leave the formatting and decoding of those bytes up to the user in the form of a script?

I think there's definitely scope for this. I wanted to use the 0x08 Modbus function before to autoscan for devices on the bus, but this is not currently possible using the Modbus read / write functions as that function code is not implemented.

Reducing these functions to taking a station address and a data field (everything between address and check fields in the below image) would work well (ignore ASCII mode I wanted to include the headers).

@gloomyandy said in Modbus Spindle Control:

That's my take on this, I'd be interested to hear @dc42 take on the matter. To help with that, do you have any feel for how often spindle commands are actually issued and what sort of polling might be needed for status reporting? Can this status polling perhaps be optimised by for instance only doing some of it immediately after issuing a command?

I would say under normal circumstances, control commands would be executed on a similar cadence to tool changes. For each machining operation in a gcode file, you would usually stop the spindle, run a tool change, set the spindle RPM and then start the spindle.

In my particular use case, I also want to be able to monitor the output speed of the VFD on a regular basis so I can pause or abort the job if the VFD triggers an error - it would be quite important to do that quickly, but I think polling the VFD every half a second or so is enough.

From a control perspective, the worst case scenario I can think of right now is my implementation of variable spindle speed control - it actively changes the rpm of the spindle up and down constantly to avoid resonance frequencies, and right now that happens at the same rate as polling the VFD (every 500ms).

From a start / stop perspective, I think the worst case might be something like the rapidchange toolchanger, which starts and stops the spindle in both directions within a couple of seconds.

So I think at the absolute worst case we're looking at half a second between spindle control commands or status checks.

Edit: I don't think status checks can be optimised to only check after commands. Imagine if you start a spindle and then run a 10 min machining op, and 8 mins in the VFD overheats and shuts down. RRF wouldn't know about that and could potentially keep running the toolpath while the spindle is stopped, which would cause all sorts of issues including potential broken tools. I do think the status checking needs to be running at all times while a spindle is running.

gloomyandy

@NineMile said in Modbus Spindle Control:

Edit: I don't think status checks can be optimised to only check after commands. Imagine if you start a spindle and then run a 10 min machining op, and 8 mins in the VFD overheats and shuts down. RRF wouldn't know about that and could potentially keep running the toolpath while the spindle is stopped, which would cause all sorts of issues including potential broken tools. I do think the status checking needs to be running at all times while a spindle is running.

As I said above I'm not a big fan of using modbus for serious errors, personally I'd say you should be using some sort of trigger from a hardware fault signal for that. But even in this case I suspect that the reality is that a second (or possibly) more would be fast enough to handle that situation.

My thoughts on checking for a response after a command was that perhaps you can use a slower polling rate during "normal operations" and only increase that rate if you need to after say issuing a command. But really that is just detail and probably depends a lot on the actual use case.

I'm curious in the table above, it lists a "start" value, but I don't see any mention of that either in the spec I linked to or the code. Where is that table from?

NineMile

@gloomyandy said in Modbus Spindle Control:

As I said above I'm not a big fan of using modbus for serious errors, personally I'd say you should be using some sort of trigger from a hardware fault signal for that. But even in this case I suspect that the reality is that a second (or possibly) more would be fast enough to handle that situation.

I agree with this to a point, but I'd take a guess that at the moment, the majority of spindles that are being run from RRF will be running without any feedback from the VFD at all - because just getting them connected up and running with direction control is complex enough, let alone working out how to get feedback inputs and writing RRF macros to process them properly.

The best thing about Modbus control is only needing to run 2 wires to the VFD, and getting 2-way communication with minimum complexity on the hardware side.

IMO It's much easier to write spindle control macros for a particular VFD type once that include some level of machine and vfd protection using Modbus feedback than it is to walk people through the hardware and VFD setup to implement hardware-based alarming and failure recovery. Especially with all the different types of VFDs out there that all seem to work slightly differently on the hardware side as well.

@gloomyandy said in Modbus Spindle Control:

I'm curious in the table above, it lists a "start" value, but I don't see any mention of that either in the spec I linked to or the code. Where is that table from?

I mentioned this briefly in the gcode I posted for the other style of control - this is from the VFD manual for the Shihlin SL3 which regardless of modbus-rtu spec, specifies a quiet time of 10ms between requests to give the VFD processor enough time to respond. If you send a request before this 10ms timer is up the request errors out.

Attaching the SL3 manual here because I think it's probably one of the most verbose VFD manuals I've come across that explains the Modbus control approach.

20220609_SL3_User_Manual_V1.03-041.pdf

NineMile

@gloomyandy said in Modbus Spindle Control:

So I think there might still be a case for having modbus specific read/write operations but I'd be tempted to try and make them more generic ones that simply read/write a bunch of bytes and apply the timing and crc to them, but leave the formatting and decoding of those bytes up to the user in the form of a script?

Looking at the Modbus gcodes again this evening (M260.1 and M261.1) and it occurs to me that if the user has the ability to send arbitrary data then there's no distinction between these 2 commands.

It seems like this lower-level "raw" behaviour could be achieved with a single gcode that takes raw bytes and expects a particular number of bytes in response, something like this:

; Purely hypothetical request that expects 3 data bytes returned from the 3 byte data request.
; var.modbusRaw would be a vector of bytes
M261.3 B3 D{10,12,14} V"modbusRaw"

The underlying implementation would send the station address and deal with CRC, but everything to do with the data would be up to the user.

dc42

@NineMile the Modbus protocol specifies what response is expected to each command. RRF checks those responses. We could ads a generic command but as the response to such a command would be undefined, we would not be able to check it.

Appending the CRC would only be appropriate if the non-Modbus device used the same CRC rules as Modbus.

gloomyandy

@dc42 I had a quick look at the linuxCNC HAL code for the VFD that the user over on discord was using. That looks like it uses the same crc16 and seed as modbus but has a different payload. I get the impression from a quick scan of the linuxCNC stuff that this is fairly common, devices which claim to be "modbus" use the modbus signal timing. framing and crc but do not use the modbus register/coil model and associated commands. But I'm certainly no expert on this.

NineMile

@gloomyandy said in Modbus Spindle Control:

@dc42 I had a quick look at the linuxCNC HAL code for the VFD that the user over on discord was using. That looks like it uses the same crc16 and seed as modbus but has a different payload. I get the impression from a quick scan of the linuxCNC stuff that this is fairly common, devices which claim to be "modbus" use the modbus signal timing. framing and crc but do not use the modbus register/coil model and associated commands. But I'm certainly no expert on this.

This is my understanding as well. Theres at least a couple quite common Chinese VFD's which advertise Modbus support but don't follow the modbus-rtu data format - the Huanyang one encountered on Discord seems to be the most common though.

For my own understanding of the RRF codebase, I spent a bit of time looking at how custom spindle control might be implemented using DoFileMacro.

It seems like adjusting spindle state and rpm would probably need to move to some type of reconciliation based system - where direct calls to spindle->SetRpm() and spindle->SetState() would store values that would then be applied later while spinning a GCodeBuffer.

Applying the settings could be simply setting pin output states for a standard spindle config, like what happens now, or calling DoFileMacro to execute a user-provided control system.

However - there's areas in the code where spindle control actions appear to be taken from outside any particular GCodeBuffer (Tool::Activate() for example, called from direct LCD requests). These would not map cleanly onto just switching into a spindle control state in the current GCodeBuffer where the request was made.

My initial thought was to use the Daemon buffer for all spindle control, but a long-running daemon macro could impact spindle control, and vice-versa.

Does it make sense to run the spindle control actions in their own GCodeBuffer instance and synchronise it with the other buffers?

For example - when File buffer calls M3 to start the spindle, we change the pending spindle settings, notify that the spindle needs to be reconciled and then don't allow the File channel to continue until the Spindle buffer reports the settings have been applied?

Any thoughts on this appreciated, or if I'm missing any prior art that might indicate a smarter way to do this?

I'm going to be at SMRRF in a couple weekends so I'd be happy to chat about this in person if you're going to be there (David / Andy)?

gloomyandy

@NineMile I'm also planning to be at SMRRF, probably on the Sunday.

dc42

@NineMile I've added new command M290.4 in 3.6-dev to support nonstandard transactions that conform only partially to the Modbus specification. See https://docs.duet3d.com/en/User_manual/Reference/Gcodes#m2604-raw-modbus-transaction. It's unlikely that I will have time to do any more work in this area before the 3.6.0 release.

gloomyandy

@NineMile I've merged and pushed David's changes to the stm32 repos. I don't currently have a modbus setup so have not been able to test either the old or new code.