I have general guidelines for preprocessing Illumina data prior to assembly. But for SARS-CoV-2 specifically, I had to make a different pipeline because the data was very nuanced, with amplification-specific artifacts. You can download BBTools and look in /bbmap/pipelines/covid/processCorona.sh, but I'll post it here for convenience. The script is a bit customized for UCSF's conventions of storing/naming files but the general process should be applicable to any artic3-amplified Illumina samples.

#!/bin/bash

##  Written by Brian Bushnell
##  Last modified April 11, 2020
##  Description:  Calls SARS-CoV-2 variants from Illumina amplicon data.
##                This script assumes input data is paired-end.

##  Usage:  processCorona.sh <prefix>
##  For example, "processCorona.sh Sample1" if the data is in Sample1.fq.gz

##  Grab the sample name from the command line.
NAME="$1"

##  Set minimum coverage for genotype calls.
##  Areas below this depth will be set to N in the consensus genome.
MINCOV=5

##  Specify the viral reference file.
##  NC_045512.fasta contains the SARS-CoV-2 genome, equivalent to bbmap/resources/Covid19_ref.fa
REF="NC_045512.fasta"

##  PCR primers
##  artic3 primers are in /bbmap/resources/artic3.fasta
PRIMERS="artic3.fasta"

##  This line is in case the script is being re-run, to clear the old output.
rm "$NAME"*.sam.gz "$NAME"*.bam "$NAME"*.bai "$NAME"*.txt "$NAME"*.fa "$NAME"*.vcf

##  If data is paired in twin files, interleave it into a single file.
##  Otherwise, skip this step.
##  In this case, the files are assumed to be named "Sample1_R1.fq.gz" and "Sample1_R2.fq.gz"
#reformat.sh in="$NAME"_R#.fq.gz out="$NAME".fq.gz

##  Split into Covid and non-Covid reads if this has not already been done.
##  This step can be skipped if non-Covid was already removed.
bbduk.sh ow -Xmx1g in="$NAME".fq.gz ref="$REF" outm="$NAME"_viral.fq.gz outu="$NAME"_nonviral.fq.gz k=25

##  Recalibrate quality scores prior to any trimming.
##  *** Requires a recalibration matrix in the working directory (see recal.sh for details). ***
##  This step is optional but useful for Illumina binned quality scores.
bbduk.sh in="$NAME"_viral.fq.gz out="$NAME"_recal.fq.gz recalibrate -Xmx1g ow

##  Discover adapter sequence for this library based on read overlap.
##  You can examine the adapters output file afterward if desired;
##  If there were too few short-insert pairs this step will fail (and you can just use the default Illumina adapters).
bbmerge.sh in="$NAME"_recal.fq.gz outa="$NAME"_adapters.fa ow reads=1m

##  Remove duplicates by sequence similarity.
##  This is more memory-efficient than dedupebymapping.
clumpify.sh in="$NAME"_recal.fq.gz out="$NAME"_clumped.fq.gz zl=9 dedupe s=2 passes=4 -Xmx31g

##  Perform adapter-trimming on the reads.
##  Also do quality trimming and filtering.
##  If desired, also do primer-trimming here by adding, e.g., 'ftl=20' to to trim the leftmost 20 bases.
##  If the prior adapter-detection step failed, use "ref=adapters"
bbduk.sh in="$NAME"_clumped.fq.gz out="$NAME"_trimmed.fq.gz minlen=60 ktrim=r k=21 mink=9 hdist=2 hdist2=1 ref="$NAME"_adapters.fa altref=adapters maq=14 qtrim=r trimq=10 maxns=0 tbo tpe ow -Xmx1g ftm=5

##  Trim artic3 primers, if present.
##  Disable this line if you are not amplifying with primers.
bbduk.sh in="$NAME"_trimmed.fq.gz out="$NAME"_trimmed2.fq.gz ref="$PRIMERS" ktrim=l restrictleft=30 k=22 hdist=3 qhdist=1 rcomp=f mm=f

##  Align reads to the reference.
##  Local flag is due to primer-amplification-induced anomalies at read ends;
##  for randomly-sheared, unamplified data, "local" should be omitted.
bbmap.sh ref="$REF" in="$NAME"_trimmed2.fq.gz outm="$NAME"_mapped.sam.gz nodisk local maxindel=500 -Xmx4g ow k=12

##  Deduplicate based on mapping coordinates.
##  Note that if you use single-ended amplicon data, you will lose most of your data here.
dedupebymapping.sh in="$NAME"_mapped.sam.gz out="$NAME"_deduped.sam.gz -Xmx31g ow

##  Remove junk reads with unsupported unique deletions; these are often chimeric.
filtersam.sh ref="$REF" ow in="$NAME"_deduped.sam.gz out="$NAME"_filtered.sam.gz mbad=1 del sub=f mbv=0 -Xmx4g

##  Remove junk reads with multiple unsupported unique substitutions; these are often junk, particularly on Novaseq.
##  This step is not essential but reduces noise.
filtersam.sh ref="$REF" ow in="$NAME"_filtered.sam.gz out="$NAME"_filtered2.sam.gz mbad=1 sub mbv=2 -Xmx4g

##  Trim soft-clipped bases.
bbduk.sh in="$NAME"_filtered2.sam.gz trimclip out="$NAME"_trimclip.sam.gz -Xmx1g ow

##  Call variants from the sam files.
##  The usebias=f/minstrandratio=0 flags are necessary due to amplicon-induced strand bias,
##  and should be removed if the data is exclusively shotgun/metagenomic or otherwise randomly fragmented.
callvariants.sh in="$NAME"_trimclip.sam.gz ref="$REF" out="$NAME"_vars.vcf -Xmx4g ow strandedcov usebias=f minstrandratio=0 maf=0.6 minreads="$MINCOV" mincov="$MINCOV" minedistmax=30 minedist=16 flagnearby

##  Calculate reduced coverage as per CallVariants defaults (ignoring outermost 5bp of reads).
pileup.sh in="$NAME"_trimclip.sam.gz basecov="$NAME"_basecov_border5.txt -Xmx4g ow border=5

##  Generate a mutant reference by applying the detected variants to the reference.
##  This is essentially the reference-guided assembly of the strain.
##  Also changes anything below depth MINCOV to N (via the mindepth flag).
##  Does not apply indels below MINCOV
applyvariants.sh in="$REF" out="$NAME"_genome.fa vcf="$NAME"_vars.vcf basecov="$NAME"_basecov_border5.txt ow mindepth="$MINCOV"

##  Make bam/bai files; requires samtools to be installed.
##  This step is only necessary for visualization, not variant-calling.
#samtools view -bShu "$NAME"_trimclip.sam.gz | samtools sort -m 2G -@ 3 - -o "$NAME"_sorted.bam
#samtools index "$NAME"_sorted.bam

##  At this point, "$NAME"_sorted.bam, "$NAME"_sorted.bam.bai, "$REF", and "$NAME"_vars.vcf can be used for visualization in IGV.

with the AmpliSeq for Illumina SARS-CoV-2 Research Panel

If you want to submit for inclusion in a certain panel or database, I would look out for their analysis guidelines and standardized workflows they might have set up and follow these closely.

Do I need fastq preprocessing? Fastq analysis looks good to me, but I might be wrong.

So you did a QC and it looked good without adapters and high quality? In principle, you don't need to do trimming and there are few adapters in modern Illumina data. However, I do it anyway to avoid having to do everything out of fear of an "em, actually,... you should have..." type of reviewer. As it is the first step of the analysis, you would have to run everything else again. But no one can complain about doing read trimming when it wasn't necessary. I'd run them through fastp, as little effort as necessary.

I have 4 forward and 4 reverse fastq files for one sample. For alignment, should I concatenate the 4 forward files into 1 and the 4 reverse files into 1 to get 2 fastq files? Is this correct?

I'd try to avoid this at all costs for several reasons and rather merge the BAM files afterward. This is better for QC and keeping all the files properly paired.

After this step, I have to obtain an assembled fasta file of the sample. To do this, should I align the two concatenated fastq files against the reference and obtain a fasta file? I believe that I might get these from bam files, but I am not certain. If this is correct, how can I get this assembled fasta file?

It depends on your research question. You can either do a de novo assembly or an alignment against a reference with variant calling. What do you need? These two options will produce different results and have different strengths and abilities to detect different kinds of variation. If you need a genome in fasta format, don't confuse a genome assembly with an ALT-genome (genome generated by inserting all alternative alleles into the reference). They are not comparable.

Is there a way to get information about amino acid substitution from VCF file?

Yes, by using tools like SnpEff with a reference annotation.