In mid-January this year, a reader contacted me regarding their set of control floppies for an Okuma CNC lathe. Owing to the age of the equipment, the control program is loaded into volatile SRAM, backed-up by a battery. Every five years, the battery needs to be replaced (or when it goes dead) and the program is lost from memory. Restoring the machine to function requires reloading this program from an initial program load (IPL) source. In the case of their lathe, this was on a set of five floppy disks (but alternatively, could be from paper tape or even audio cassette). Unfortunately for them, the disks proved to be unreadable.
After a few back and forth messages, I established that the disks were likely to be physically recorded in MFM format with IBM-style geometry on High Density (HD) media, but with a custom filesystem that needed the use of a custom program to handle the files on the disk. This is something the Kryoflux (and indeed, most regular floppy drives) can handle in regards to making direct images and writing a copy. The biggest challenge was the fact the media was physically in bad shape, and it seemed highly unlikely to me that a full recovery would be possible.
Despite explaining this, the reader still wished for me to give them a try with the Kryoflux to see if anything could be recovered from them, as there was nothing to be lost. I agreed, on the condition that postage costs are covered both ways, and that I would not accept any payment (owing to the licensing conditions of the Kryoflux).
The Journey Begins
Thanks to Australia Post, the disks arrived safely the next day.
As instructed, the disks were packaged safely, with some level of mechanical protection and physical separation to prevent any further damage to the disks.
The full set of five disks were sent, two of which (Disk 1 and 5) were not functioning. The other three were claimed to be working.
A look at a working Disk 3 was already not encouraging. A look at the media through light shows the oxide coat for one track clearly being “carved” off the film base by a read head in the past. It’s almost certain that the data is lost in that track, so the fact the disk works is probably a pure miracle (e.g. no data recorded at that location). At least, I have been reassured that the disk was successfully imaged in the past, so the physical condition is probably of little concern for this disk.
A look at Disk 2 which was also claimed to be working shows both irregular scoring of the surface which suggests the head may have been “bouncing” along the track with a partial head clog (as oxide is abrasive and will damage the media surface). Also visible are concentric rings where the media has been scored, but not sufficiently to cause the oxide to lift. Owing to the condition of these “good” disks, I wasn’t too optimistic about the bad ones. I decided not to even attempt reading these disks to avoid further damage to the disks or to my equipment, and instead looked at the image files to see if anything interesting could be gleaned. As far as I can tell, it has its own relatively primitive filesystem and format that has some “magic bytes” scattered around the beginning of each file, but without knowing more about the machine, I couldn’t make any further deductions.
The first of the bad disks was Disk 1. This one showed much of the same, but instead of being limited to one track, there was a range of sequential tracks where the oxide had been lost completely. Not very encouraging.
The other bad disk was Disk 5, and it’s in arguably worse shape with even more oxide loss. I have not experienced such dramatic oxide loss of high density 3.5″ floppy disks until this case. I’ve experienced similar with double density disks, but it just goes to show that if you have any data on floppy disks, it’s probably a good time to image it to other storage ASAP or risk losing it.
The disks themselves are unbranded, with a flexible black plastic shutter. These seem quite similar (superficially) to some I had bought in bulk from a duplication house in Hong Kong around 2000. I don’t have any of those anymore, as they didn’t seem to be of high quality and did cause problems – so I can’t rule out the possibility that Okuma had used low quality media to hold the IPL program. This isn’t helped by the fact that many industrial environments are quite harsh in terms of temperature, humidity and chemical atmosphere, so that may have accelerated the demise of the disks. Add to this the fact the disks are serial-number married to their respective machines, that IPL only needs to be performed once in a few years and the urgency of backing the data up doesn’t seem to be there. On the other hand, it is said that Okuma makes a lot of money from providing replacement disks … so it’s probably a good idea to make a backup as soon as you can.
Giving it a Shot
Seeing that the disks made it all the way to my house, regardless of how pessimistic I was about their chances, I would have to give them a go to be absolutely sure. I was hoping that there would be other very similar disk sets where any “missing” tracks might be able to be filled in with the data from their images to make a working disk set. But even sourcing any other images proved to be hard enough.
I have heard of oxide shedding from reel to reel tape due to sticky shed syndrome. To remedy this temporarily, a reel can be baked at a low heat to drive out the moisture to temporarily restore the binder to functionality. I decided I would experiment with a similar idea for this set of disks in case it improved the situation. I built a quick shelf of cardboard over my PC’s top exhaust fans with ~80mm holes to expose the 3.5″ disk shell area. As I was running an intensive GPU workload, the heat output was moderate and by adjusting the height of the cardboard and monitoring a thermocouple, I was able to achieve about 45C for a period of 12 hours before attempting to read the disks.
The reading strategy needed a bit of thought. The Kryoflux likes to run a good number of retries when it can’t read any data, however, dwelling on a flaky track is likely to damage it quite severely. This is relatively minor, compared to the possibility that the dislodged oxide might lodge in the read head and then go on to damage the next track the Kryoflux attempts to read. As a result, the first step was to change the retry setting to a very timid number of just 3 retries. The read mode was set to guided STREAM (but non-preservation) so as to limit the number of revolutions per track. After each read, the drive was cleaned and the disk retried. After three reads, I swapped to a different brand drive to see if more data could be recovered.
The results of five reads of Disk 1 were not encouraging. The most important areas of the disk are generally the outer tracks (lower track numbers) as that is where the recording starts from and filesystem metadata are usually stored. In this case, many of the tracks on Side 1 were unidentifiable. One drive did seem to do better than the other, but only the tracks that come up with orange-H or green blocks are completely recovered. This leaves a significant area of the disk unrecovered, and thus even writing a partial-copy disk doesn’t seem to be worthwhile.
A Look at the STREAMs
A few of the tracks come up good. A good MFM recording has three “bands” with a clear separation between them. There is some peak-shift which is probably down to the quality of the media and the drive used to initially record the disk.
Some of the tracks with lost data seem to show either a lot of peak-shifting which suggests a weakening of the signal on the disk to the point that the “margin” between different signal periods is not clearly distinguishable.
Some tracks seem to have a pattern which suggests that half of the disk had somehow been magnetically weakened. This may be because a weak magnet had run through that area of the disk, maybe it was sitting near a CRT monitor degauss coil at a distance or maybe a power cable.
Other tracks had a more uniform weakening resulting in the symbol periods drifting wildly and occasionally moving into “noise”.
One side of one disk had a distinctive radial section at the middle of each revolution where the signal turned noisy. This implies a radial damage, consistent relative to the index signal. When I asked the reader about this, they admitted to cleaning off some gunk from the disk surface using a soft cloth and some methylated spirits. Unfortunately, this is likely to have caused permanent damage, as ethanol does dissolve the oxide binders and can “smear” the oxide around damaging the recording. It’s part of the reason we use it to clean heads.
This one shows mostly noise, and is what comes from the tracks where the oxide had almost been completely lost. The data isn’t there.
Rather unfortunately, but also rather predictably, these disks were not recoverable. I can’t say for sure whether the baking process helped or hindered the recovery, but even just looking at the physical condition of the disks as received, it seems it was never meant to be.
The disks themselves may have been low quality media, stored in sub-optimal conditions, but it is a reminder that even High Density 3.5″ floppy disk media can be victim of oxide shedding and that recovery of data from floppy disks is an urgent priority to prevent data loss.
In the case of these floppy disks, these flimsy pieces of plastic and metal are basically all that stand between a fairly expensive machine working and not working, so it’s rather disappointing to have not been able to have better news to offer.